[TianJialong]

What changes are proposed in this pull request?

Fixes #13471

Maps the temporal string/number conversion expressions to their velox flinksql implementations (companion velox PR bigo-sg/velox#94, velox4j PR bigo-sg/velox4j#75):

• RexCallConverterFactory: TO_DATE / TO_TIMESTAMP / TO_TIMESTAMP_LTZ entries — plain name mappings; both arities of the first two are covered by the flinksql signatures, TO_TIMESTAMP_LTZ's numeric argument types bind through the same registration.
• RexNodeConverter: DATE literals convert to the epoch-day IntegerValue — Flink folds constant TO_DATE calls in the planner, and the plan used to die at conversion with the DATE literal unsupported. TIMESTAMP_WITH_LOCAL_TIME_ZONE literals share the TIMESTAMP conversion body (TimestampString.getMillisSinceEpoch → boxed-Long millis, with a null guard), so folded TO_TIMESTAMP / TO_TIMESTAMP_LTZ literals reach velox. NULL literal branches emit a null variant (a velox4j serialization limitation currently drops typed NULL literals at the bridge; NULL rows arrive from data inputs instead).
• RexNodeConverter: identity-cast elimination — the planner emits CAST(literal AS T) around every folded literal whose nullability differs from the sink's, and velox rejects same-type casts, so isSameTypeIgnoringNullability (recursive type equality ignoring nullability only; ANY never elided; composite component types must match) passes the operand through unchanged.
• RexNodeConverterTest: covers the two literal channels, the identity-cast behavior across scalar and composite types, and the NULL literal branches.

How was this patch tested?

• Planner-level: RexNodeConverterTest — DATE literal converts to the epoch-day IntegerValue, LTZ literal to epoch millis; identity CAST elided on nullability-only casts for scalar and composite types (ARRAY/ROW/MAP casts with different component types stay real casts); NULL literal branches.
• Execution-level (embedded gluten stack): DateTimeFunctionsTest in org.apache.gluten.table.runtime.stream.custom — both arities of to_date (incl. constant-format folding through the DATE literal channel), to_timestamp (SMART clamping, 24:00 rollover, fraction flooring, column/constant formats), to_timestamp_ltz (precision 0/3 unit semantics, epoch-domain bounds, double/decimal inputs), NULL propagation on all forms, and CASE/WHERE combinations.
• End-to-end dual-run on Flink 1.19.2: 32 SQL scenarios across the family — both arities, constant vs column formats, NULL propagation, literal folding for DATE and LTZ, invalid-input and out-of-range classes — with 29 exact matches and 3 shared job-failure classes (illegal format, year overflow, unsupported LTZ precision: identical failures on both stacks), zero mismatches.

[TianJialong]

@KevinyhZou @lgbo-ustc @philo-he @ggjh-159 Would you mind reviewing this PR?

[lgbo-ustc]

This PR depends on bigo-sg/velox#94, but the Flink build still pins a Velox4j revision whose Velox dependency predates those functions. Please update the Velox and Velox4j references so the new offload mappings are built and tested against a version that contains the required implementations.

[lgbo-ustc]

[P1] These new TO_DATE / TO_TIMESTAMP / TO_TIMESTAMP_LTZ mappings require the flinksql registrations introduced by bigo-sg/velox#94. However, this PR does not change .github/workflows/flink.yml, which still checks out bigo-sg/velox4j at 26c7715278e6f6795084f6334998cb2ce382f7aa (lines 90-92). That Velox4j revision reads src/main/cpp/velox-ref.txt and pins Velox to aff6baab84d650cc22dd8b7cf1ab55f0ea52d929, which is 22 commits behind the #94 head (6aa6b617f45365bde843dcdf4546a7eb602e2f17) and has no TemporalFunctionsRegistration.cpp. Please update the Velox4j Velox ref to a commit containing #94, publish the corresponding Velox4j revision, and pin that Velox4j revision in this PRs Flink workflow. Otherwise the new planner expressions will reach a native engine without these functions and fail at execution; the current CI build cannot validate the feature.

[TianJialong]

@lgbo-ustc Thanks for the review! Pushed 157cb39 (Additionally, some test cases were added), please check.