…r booleans

History::count() and History::search() now clear more_info_json before
calling construct_query, matching the documented contract that non-typed
trait methods do not evaluate JSON filters.

For more_info_json filter values of "true" or "false", use SQLite's
json_type() instead of CAST(json_extract(...) AS TEXT) to match JSON
boolean values exactly. This prevents false matches against integer 1/0
and string "true"/"false". Other filter values continue to use CAST.

…alue enum

Add `JsonFilterValue` enum with `Null`, `Bool`, `Integer`, `Real`, and `Text`
variants so that `SearchFilter::more_info_json` filters are type-precise.
Each variant generates SQL using `json_type()` to prevent collisions between
JSON booleans, integers, and strings (e.g. `Bool(true)` no longer matches
integer `1` or string `"true"`).

Export `JsonFilterValue` from the crate root. Add collision tests covering
every scalar type combination.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Add json_filter_value_real_vs_integer to verify Real(f) does not collide
with Integer values sharing the same numeric representation, and
json_filter_value_null_vs_missing_path to verify Null matches JSON null
at a path but not SQL-NULL more_info or an absent path.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

json_filter_value_real_vs_integer: add an integer row at $.score to verify
Real(1.0) excludes it (json_type 'integer' != 'real') while Integer(1)
correctly matches it but not the real rows.

json_filter_value_null_vs_missing_path: add a row whose non-null JSON
has no $.val key (WithoutVal struct) to verify JsonFilterValue::Null
excludes it alongside the SQL-NULL more_info row.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

History::save now delegates to save_with_extra rather than a private
save_impl, removing the unnecessary indirection layer.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

History::update() went through the type-erased load/save path, which
always deserializes more_info as IgnoreAllExtraInfo and re-serializes
it as null, silently wiping out data written via save_with_extra. Any
consumer that calls Reedline::update_last_command_context for normal
post-command bookkeeping would clobber typed more_info on every run.

Change the trait update() to update every column except more_info,
and add an update_with_extra<E>() inherent method for callers that
need to read/modify typed more_info atomically.

Also mark JsonFilterValue as non_exhaustive since array/object
support is a plausible future addition.

Add update_with_extra_blocks_concurrent_writer, which uses two file-backed
connections to the same database to confirm that update_with_extra's
immediate transaction actually blocks a concurrent writer (previously there
was no test that would fail without the transaction).