Speed up the Parameter get/set hot paths

[philippjfr]

Summary

Parameter.__get__ and Parameter.__set__ are the hottest code in param: every
attribute read and write on a Parameterized instance goes through them, and
downstream libraries like Panel do that millions of times in a session. This PR
adds asv coverage for those paths (there was none), then removes two sources of
redundant work that the coverage exposed.

Reads of a Number parameter are now roughly 2x faster. Reads of other
parameter types improve by 17-38%, and writes by 3-14%. No behaviour changes, no
API changes, no new slots, and the full test suite passes unchanged
(1514 passed, 4 skipped, 2 xfailed).

Changes

bench: Add asv benchmarks for parameter value get/set hot paths

The existing suite covered class creation, .param access, depends and watcher
triggering, but had no benchmark for plain parameter value get/set. Adds four
suites: ParameterizedGetValueSuite and ParameterizedSetValueSuite
(parameterized over Parameter, Number, String, Boolean, at both class and
instance level), ParameterizedFirstSetValueSuite (the one-off cost of the first
set, which triggers the per-instance Parameter copy), and
ParameterUnboundSlotSuite (slot reads on an unbound Parameter, with and without
_slot_defaults fallback).

The get/set suites warm up in setup so they measure the steady state rather
than first-touch instantiation.

perf: Avoid redundant Parameter slot reads in __get__ and __set__

Slot reads on a Parameter are not free. Parameter.__getattribute__ is
overridden (parameterized.py:1888) to resolve Undefined slots against
_slot_defaults, which costs about 55ns per read instead of the ~13ns an
unoverridden attribute read would cost, because a Python-level
__getattribute__ defeats CPython's adaptive LOAD_ATTR specialization.

__get__ read self.name twice and __set__ read it five times. Both now bind
it to a local once. __set__ also read self.default eagerly even on the common
path where a stored value already exists, and re-read obj._param__private
repeatedly; the default read now happens only in the KeyError branch where
there is no stored value, and _param__private is bound to a local.

perf: Resolve Number value and dynamism in a single pass

Number.__get__ called super().__get__() to resolve the value, then
_value_is_dynamic() to decide whether to re-validate, and that second call
walked the entire descriptor chain a second time. There was already a maintainer
TODO on the method saying as much.

Adds Dynamic._resolve_dynamic(), which returns (value, is_dynamic) from a
single traversal. Dynamic.__get__ is now a thin wrapper over it, and
Number.__get__ uses it directly. The dynamism probe also checks callable(gen)
before hasattr(gen, '_Dynamic_last'), since generators are always callable and
the cheap check short-circuits the common static-value case.

_value_is_dynamic is untouched and still used by its other callers.

Benchmarks

asv on macOS / arm64, each commit built and run in its own isolated
environment. Baseline is the commit that adds the benchmarks.

Reads:

Benchmark	Before	After	Ratio
GetValue.time_instance('Number')	688±50ns	334±20ns	0.49
GetValue.time_class('Number')	709±20ns	417±50ns	0.59
GetValue.time_instance('Boolean')	271±20ns	167±20ns	~0.62
GetValue.time_instance('Parameter')	250±30ns	208±20ns	~0.83
GetValue.time_instance('String')	250±20ns	209±40ns	~0.83
UnboundSlot.time_undefined_slot	250±20ns	208±20ns	~0.83

Writes:

Benchmark	Before	After	Ratio
SetValue.time_instance('Parameter')	1.52±0.04μs	1.31±0.04μs	~0.86
SetValue.time_instance('String')	1.58±0.1μs	1.38±0.02μs	~0.87
SetValue.time_instance('Boolean')	1.42±0.02μs	1.35±0.04μs	0.96
SetValue.time_instance('Number')	2.02±0.06μs	1.96±0.04μs	0.97
SetValue.time_class('Number')	3.75±0.1μs	3.58±0.06μs	0.96
WatcherSuite.time_trigger	9.73±0.2μs	9.25±0.2μs	0.95

Attribution: the slot-read commit is the broad but shallow win (every read
0.77-0.83, Number reads 688 to 563ns, set('Parameter') 1.52 to 1.31μs). The
_resolve_dynamic commit is where the Number halving comes from (563 to 334ns
instance, 688 to 417ns class).

No regressions. ParameterizedParamAccessSuite.time_class reports 125 to 166ns
in the second step, but that operation measures 28.6ns in-process, well below
asv's ~42ns timer granularity on this machine, and it was flat across the first
step. That granularity is also why so many sub-microsecond figures land on
exactly 83/125/167/208/250ns; the Number read result is the one sub-microsecond
number comfortably outside it, and it is corroborated in-process at 671 to 206ns.

AI Disclosure

PR created with heavy help from Claude Opus 5.

[codecov]

Codecov Report

❌ Patch coverage is 97.61905% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 86.75%. Comparing base (bd43b28) to head (c9d028e).

Files with missing lines	Patch %	Lines
param/parameterized.py	97.05%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1167      +/-   ##
==========================================
+ Coverage   86.73%   86.75%   +0.02%     
==========================================
  Files           9        9              
  Lines        5321     5330       +9     
==========================================
+ Hits         4615     4624       +9     
  Misses        706      706              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[philippjfr]

Full asv benchmark output comparing before (the commit adding more benchmarks) and the final commit:

| Change   | Before [bc93512a] <perf-hot-paths~3>   | After [3bee86e5] <perf-hot-paths>   | Ratio   | Benchmark (Parameter)                                                                      |
|----------|----------------------------------------|-------------------------------------|---------|--------------------------------------------------------------------------------------------|
|          | 22.0±0.6ms                             | 22.2±0.6ms                          | 1.01    | benchmarks.ImportSuite.timeraw_import_param                                                |
|          | 826±7ns                                | 835±20ns                            | 1.01    | benchmarks.ParameterSuite.time_instantiation                                               |
|          | 104±20ns                               | 83.4±0.2ns                          | ~0.80   | benchmarks.ParameterUnboundSlotSuite.time_concrete_slot                                    |
|          | 208±0.5ns                              | 208±20ns                            | 1.00    | benchmarks.ParameterUnboundSlotSuite.time_undefined_slot                                   |
|          | 7.81±2μs                               | 8.31±1μs                            | 1.06    | benchmarks.ParameterizedDependsInstantiateSuite.time_declarative_10_parameters_separate_cb |
|          | 8.46±0.9μs                             | 8.63±0.5μs                          | 1.02    | benchmarks.ParameterizedDependsInstantiateSuite.time_declarative_10_parameters_shared_cb   |
|          | 7.94±0.4μs                             | 8.65±0.9μs                          | 1.09    | benchmarks.ParameterizedDependsInstantiateSuite.time_declarative_1_parameter               |
|          | 144±3μs                                | 146±5μs                             | 1.01    | benchmarks.ParameterizedDependsInstantiateSuite.time_watch_10_parameters_separate_cb       |
|          | 114±4μs                                | 116±2μs                             | 1.01    | benchmarks.ParameterizedDependsInstantiateSuite.time_watch_10_parameters_shared_cb         |
|          | 33.2±2μs                               | 31.9±1μs                            | 0.96    | benchmarks.ParameterizedDependsInstantiateSuite.time_watch_1_parameter                     |
|          | 302±4μs                                | 305±2μs                             | 1.01    | benchmarks.ParameterizedDependsSuite.time_declarative_10_parameters_separate_cb            |
|          | 271±3μs                                | 277±5μs                             | 1.02    | benchmarks.ParameterizedDependsSuite.time_declarative_10_parameters_shared_cb              |
|          | 45.9±1μs                               | 44.6±0.2μs                          | 0.97    | benchmarks.ParameterizedDependsSuite.time_declarative_1_parameter                          |
| -        | 352±20μs                               | 308±4μs                             | 0.87    | benchmarks.ParameterizedDependsSuite.time_watch_10_parameters_separate_cb                  |
|          | 282±10μs                               | 273±2μs                             | 0.97    | benchmarks.ParameterizedDependsSuite.time_watch_10_parameters_shared_cb                    |
|          | 47.0±0.4μs                             | 45.8±0.3μs                          | 0.97    | benchmarks.ParameterizedDependsSuite.time_watch_1_parameter                                |
|          | 24.2±0.8μs                             | 22.7±0.6μs                          | 0.94    | benchmarks.ParameterizedFirstSetValueSuite.time_instantiate_and_first_set                  |
|          | 250±0ns                                | 250±20ns                            | 1.00    | benchmarks.ParameterizedGetValueSuite.time_class('Boolean')                                |
| -        | 708±0.2ns                              | 437±20ns                            | 0.62    | benchmarks.ParameterizedGetValueSuite.time_class('Number')                                 |
|          | 250±20ns                               | 250±0ns                             | 1.00    | benchmarks.ParameterizedGetValueSuite.time_class('Parameter')                              |
|          | 271±40ns                               | 291±20ns                            | 1.08    | benchmarks.ParameterizedGetValueSuite.time_class('String')                                 |
|          | 250±20ns                               | 250±20ns                            | 1.00    | benchmarks.ParameterizedGetValueSuite.time_instance('Boolean')                             |
| -        | 729±20ns                               | 375±30ns                            | 0.51    | benchmarks.ParameterizedGetValueSuite.time_instance('Number')                              |
|          | 250±20ns                               | 209±20ns                            | ~0.84   | benchmarks.ParameterizedGetValueSuite.time_instance('Parameter')                           |
|          | 270±20ns                               | 250±40ns                            | 0.92    | benchmarks.ParameterizedGetValueSuite.time_instance('String')                              |
|          | 26.1±2μs                               | 28.2±2μs                            | 1.08    | benchmarks.ParameterizedInstantiateSuite.time_100_parameters                               |
|          | 11.6±0.7μs                             | 13.6±4μs                            | ~1.17   | benchmarks.ParameterizedInstantiateSuite.time_10_parameters                                |
|          | 10.8±2μs                               | 10.4±2μs                            | 0.97    | benchmarks.ParameterizedInstantiateSuite.time_1_parameters                                 |
|          | 83.1±20ns                              | 125±20ns                            | ~1.50   | benchmarks.ParameterizedParamAccessSuite.time_class                                        |
|          | 229±20ns                               | 208±0.6ns                           | ~0.91   | benchmarks.ParameterizedParamAccessSuite.time_instance                                     |
|          | 291±30ns                               | 250±20ns                            | ~0.86   | benchmarks.ParameterizedParamContainsSuite.time_class                                      |
|          | 416±20ns                               | 375±20ns                            | ~0.90   | benchmarks.ParameterizedParamContainsSuite.time_instance                                   |
|          | 2.27±0.1μs                             | 2.25±0.04μs                         | 0.99    | benchmarks.ParameterizedSetValueSuite.time_class('Boolean')                                |
|          | 3.63±0.08μs                            | 3.77±0.1μs                          | 1.04    | benchmarks.ParameterizedSetValueSuite.time_class('Number')                                 |
|          | 2.21±0.04μs                            | 2.08±0.08μs                         | 0.94    | benchmarks.ParameterizedSetValueSuite.time_class('Parameter')                              |
|          | 2.17±0.02μs                            | 2.19±0.09μs                         | 1.01    | benchmarks.ParameterizedSetValueSuite.time_class('String')                                 |
|          | 1.46±0.07μs                            | 1.40±0.07μs                         | 0.96    | benchmarks.ParameterizedSetValueSuite.time_instance('Boolean')                             |
|          | 2.10±0.07μs                            | 1.98±0.04μs                         | 0.94    | benchmarks.ParameterizedSetValueSuite.time_instance('Number')                              |
|          | 1.44±0.04μs                            | 1.42±0.05μs                         | 0.99    | benchmarks.ParameterizedSetValueSuite.time_instance('Parameter')                           |
|          | 1.60±0.1μs                             | 1.48±0.06μs                         | 0.92    | benchmarks.ParameterizedSetValueSuite.time_instance('String')                              |
|          | 2.10±0.04μs                            | 2.08±0.1μs                          | 0.99    | benchmarks.ParameterizedSetattrSuite.time_class                                            |
|          | 12.9±0.3μs                             | 12.8±0.6μs                          | 1.00    | benchmarks.ParameterizedSetattrSuite.time_instance                                         |
|          | 11.9±0.2μs                             | 12.0±0.3μs                          | 1.01    | benchmarks.ParameterizedSuite.time_class_bare                                              |
|          | 2.29±0.01ms                            | 2.35±0.03ms                         | 1.02    | benchmarks.ParameterizedSuite.time_class_with_100_parameter                                |
|          | 246±2μs                                | 244±20μs                            | 0.99    | benchmarks.ParameterizedSuite.time_class_with_10_parameter                                 |
|          | 38.0±7μs                               | 37.4±1μs                            | 0.98    | benchmarks.ParameterizedSuite.time_class_with_1_parameter                                  |
|          | 9.50±0.4μs                             | 10.4±0.4μs                          | 1.10    | benchmarks.WatcherSuite.time_trigger                                                       |