Replace finders inheritance hierarchy with composition - #16154

- Fix a dangling Javadoc cross-reference in SingleResultFinderSpec to
  a nonexistent "field-ordering note".
- Add an explicit 0 * query.projections() assertion in
  RxListResultFinderSpec so the "no distinct()" quirk documented in
  RxListResultFinder's class Javadoc is self-verifying rather than
  incidental.
- Add a createFinderInvocation-level test for the And/Or literal-split
  quirk in DynamicFinderSpec; previously only buildMatchSpec (the
  compile-time matcher) exercised it, leaving the actual runtime path
  finders call unverified.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

FinderMethod declares invoke(Class, String, ...) with raw Class/
Closure parameters (suppressed at the interface level), so every
override necessarily reintroduces the same raw-type warning at its
own call site - IntelliJ flags each override individually, and
ListOrderByFinder already suppresses per-method for the same reason.
Apply the same @SuppressWarnings("rawtypes") to CountFinder's three
invoke overloads, including the DetachedCriteria one, which stays raw
to match DynamicFinderInvocation.setDetachedCriteria's existing raw
parameter type rather than diverging from it.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

- operatorPatterns/operators fields are only ever assigned in the
  constructor - make them final.
- Diamond operator for the methodExpressions map initializer.
- Merge two identical-body SecurityException/NoSuchMethodException
  catch branches into multi-catch, in both the static initializer and
  registerNewMethodExpression.
- Suppress the intentionally-ignored Matcher.find() result in
  createFinderInvocation (isMethodMatch already guarantees the match
  succeeds before a finder's invoke() is ever called).
- buildMatchSpec: the local pattern "(prefix)([A-Z]\w*)" has no
  alternation, so group 2 is guaranteed non-null whenever find()
  succeeds - drop the always-true querySequence != null conjunct, and
  collapse the now childless outer if into the inner one.
- Drop two self-reassignments (x = getInitializedExpression(x, args))
  now that the helper is confirmed to always mutate-and-return the
  same reference; and delete a soloArgs array that was populated but
  never read.
- Extract the ~25-line fetch/cache argument-map handling duplicated
  verbatim between both populateArgumentsForCriteria overloads into a
  shared applyFetchAndCacheArguments helper, parameterized over the
  join/cache operations since Query and BuildableCriteria don't share
  a supertype exposing them. Convert the surviving sort-map branches
  to pattern variables, dropping an unused `value` local along the
  way.
- configureQueryWithArguments: fold the instanceof check and cast into
  a single pattern-variable condition.
- Drop the redundant Iterable cast on methodExpressions.keySet() and
  replace criteria.remove(0) with removeFirst().
- Leave targetClass's unused-parameter warning suppressed rather than
  removing the parameter (call-site compatibility with
  grails-data-hibernate5/7 and grails-data-mongodb) and leave the
  Class<? extends Object> wildcard alone (documented GROOVY-9460
  workaround already noted in the existing TODO).

Verified via the full grails-datamapping-core and
grails-datamapping-core-test suites - no behavior change.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Checked: GROOVY-9460 (Class<D> rejected against a Class<?> parameter
under static compilation) was fixed in Groovy 2.5.16/3.0.6/4.0.0-alpha-1.
This project is on Groovy 5.0.x, well past the fix, so the
Class<? extends Object> workaround is obsolete - simplify to Class<?>.

Verified against every real call site, including the @CompileStatic
grails.gorm.DetachedCriteria#list, which passes a Class<T> field
(exactly the pattern the bug affected): grails-datamapping-core,
grails-datamapping-rx, and grails-data-hibernate5-core all compile
clean, and the core test suite is green.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

- getInitializedExpression's return value is now unused at both call
  sites (the last commit dropped the self-reassignments) - change it
  to void and drop the stale "@return"/removed-IsNull-path Javadoc
  that no longer matched the commented-out code.
- Both populateArgumentsForCriteria overloads wrapped their sort-object
  instanceof chain in a redundant "if (sortObject != null)" guard -
  instanceof already returns false for null, so the wrapper is
  eliminable without changing behavior.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Both the "select"/LAZY-named-mode if-branch and the final fallback
returned FetchType.LAZY, so the if condition was a no-op guard around
an outcome identical to just falling through - drop it.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

javaClass/methodName/arguments/expressions/criteria/operator are only
ever set in the constructor and have no setters - make them final.
detachedCriteria keeps its mutable setter, so it stays as-is.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

createFinderInvocation's Class/Closure parameters are raw to match
FinderMethod's existing raw signature (finder invoke() overloads
delegate straight through). Suppress at the interface level, same as
FinderMethod itself already does.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

- match.find() result is intentionally ignored - isMethodMatch already
  guarantees the match succeeds before invoke() is ever called, same
  as DynamicFinder#createFinderInvocation.
- new LinkedHashMap((Map) arguments[0]) is an unchecked copy-constructor
  call on a raw Map, needed since FinderMethod's invoke() signature is
  itself raw.

Both suppressed on the enclosing invoke() method, alongside the
existing rawtypes suppression.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

- Raw Class/Closure/DetachedCriteria on the three invoke overloads,
  same pattern as CountFinder/SingleResultFinder.
- findAllByBoolean(MappingContext) had no caller anywhere - it mirrors
  the Datastore/MappingContext factory-pair pattern every finder class
  uses (the MappingContext overload exists for unit testing without a
  full Datastore, as findAllBy(MappingContext) already is), so add the
  missing test instead of removing the factory.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

- Raw Class/Closure/DetachedCriteria on the three invoke overloads.
- constructFromEqualExpressions' Map.put on the raw HashMap is
  "unchecked", a separate category from the "rawtypes" it already
  suppressed.
- findByBoolean/findOrCreateBy/findOrSaveBy(MappingContext) had no
  callers, same Datastore/MappingContext factory-pair pattern as
  ListResultFinder - add the missing stateless-mode tests (mirroring
  findBy(MappingContext)'s existing one) instead of removing them.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Every flagged constructor is genuinely exercised, just never via a
literal "new Xxx(...)" expression for these particular classes:
DynamicFinder's registry invokes the (Class, String) constructor
reflectively via Constructor.newInstance, and MethodExpressionSpec's
existing round-trip tests invoke both shapes reflectively via
Class.getConstructor(...).newInstance(...) - so IntelliJ's static
usage tracker can't see either call site. Suppress rather than add
redundant tests, since coverage already exists.

Also suppress the base constructor's unused targetClass parameter -
the field it feeds is already @deprecated and unused by design, kept
only so every subclass can offer the (Class, String) shape uniformly.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

convertArguments and convertArgumentsForProp each had their own copy
of "fall back to the identity property when getPropertyByName returns
null" - extract a shared resolveProperty helper instead. This also
lets convertArgumentsForProp drop its now-unused PersistentEntity/
propertyName parameters, and removes NotInList#convertArguments'
redundant self-assigned locals along the way.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

…ug in RxSingleResultFinder

Adds tests for setPattern delegation, the null-detachedCriteria branch of the
reflective 4-arg invoke() overload, the ConversionException->MissingMethodException
mapping, DynamicFinder's buildMatchSpec/configureQueryWithArguments/
populateArgumentsForCriteria branches, and ListOrderByFinder's Map-argument
handling, across the six composed finder classes (sync + rx).

Also fixes a genuine latent bug surfaced while closing these gaps:
RxSingleResultFinder's findOrCreateBy*/findOrSaveBy* empty-result path ran
inside a spawned Thread with no exception handling, so a thrown
MissingMethodException (e.g. from a non-Equal expression) killed the thread
silently instead of reaching the Observable's error channel - any blocking
caller would hang forever. Wrapped the thread body so failures now propagate
via Subscriber#onError, and added a test proving the fix.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>