chore(audit): audit cast across Spark 3.4.3, 3.5.8, 4.0.1, 4.1.1

[andygrove]

Which issue does this PR close?

Closes #.

Rationale for this change

Completes the per-category expression audit started in earlier PRs (#4469 struct, #4470 json, #4473 collection, #4474 misc, #4475 conditional, #4476 hash, #4478 map, #4479 bitwise, #4480 predicate, #4483 array, #4486 math), using the updated audit-comet-expression skill in #4468.

cast is the only expression in conversion_funcs; the other entries (bigint, binary, boolean, ...) are SQL-syntactic shorthands for CAST(... AS X) and resolve to Cast.

What changes are included in this PR?

Support-doc audit notes

Add a per-version audit sub-bullet to cast in docs/source/contributor-guide/spark_expressions_support.md. Highlights:

• Cast(child, dataType, timeZoneId, evalMode) signature is stable across 3.4 / 3.5 / 4.0 / 4.1; per-version EvalMode resolution lives in CometExprShim.
• Comet uses a per-source-type support matrix in CometCast.scala that returns Compatible / Incompatible(reason) / Unsupported(reason) for each (from, to) pair.
• Spark's numeric-to-numeric canCast matrix is unchanged across all four versions; Spark 4.0 adds VariantType and collated StringType; Spark 4.1 adds TimeType and geospatial types with their own arms. None of those new arms are in CometCast, so they fall back to Spark.

Support-level consistency fixes

None in this PR. The audit surfaced cosmetic inconsistencies (e.g. canCastFromFloat/Double/Decimal ignore evalMode; the static getUnsupportedReasons / getIncompatibleReasons don't enumerate the per-pair reasons returned by isSupported; one case _: Incompatible() with no reason in the array-of-binary branch) but these touch a load-bearing file and are deferred to follow-ups.

Tracking issues filed for follow-up

• [Bug] CAST(BinaryType AS StringType) uses unsafe from_utf8_unchecked (undefined behaviour) #4488 CAST(<binary> AS STRING) uses unsafe String::from_utf8_unchecked (undefined behaviour for non-UTF8 inputs).
• [Doc] CAST collated-string handling on Spark 4.0+ is implicit and untested #4489 Spark 4.0 collated-string casts are implicitly unhandled; no test asserts the fallback (umbrella [Spark 4.0] Add string collation support #2190).
• [Doc] CAST has no explicit TimeType branch (Spark 4.1) #4490 Spark 4.1 TimeType casts have no explicit Unsupported arm; falls back implicitly but doesn't appear in the auto-generated compat doc.
• [Feature] CAST(MapType AS MapType) falls back even though native cast_map_to_map exists #4491 CAST(<map> AS <map>) falls back even though native cast_map_to_map exists.
• [Bug] CAST(complex AS STRING) does not honour spark.sql.legacy.castComplexTypesToString.enabled #4492 spark.sql.legacy.castComplexTypesToString.enabled=true is not honoured.

Existing #1371 (CAST(float|double AS decimal) rounding) and #2190 (Spark 4.0 collation umbrella) are also referenced from the doc sub-bullet.

Audit process

Audited using the audit-comet-expression skill (4 Spark versions per #4468), driven by 3 parallel agents covering high-level + numeric, string conversions, and datetime + complex types.

How are these changes tested?

• make core succeeds (no code changes; doc only).
• Existing test coverage in CometCastSuite and the cast-related SQL fixtures remains unchanged.