[Feature] Support TIME in Lance file format #477

Search before asking

• I searched in the issues and found nothing similar.

Motivation

Paimon Java accepts TIME columns in Lance data files: LanceFileFormat.LanceRowTypeVisitor.visit(TimeType) accepts the type, ArrowFieldTypeConversion writes it as Arrow TIMEMILLI (time32[ms]), and the Lance type mapping in Paimon's docs/docs/concepts/spec/fileformat.md lists TIME → TIME32 / TIME64.

Paimon C++ has represented TIME as time32[ms] since #378, and Parquet and Vortex already support it. Lance does not:

• SchemaValidation::ValidateLanceDataField() has no TIME32 case. A TIME column falls through to default and is rejected with Lance file format does not support type time32[ms] (src/paimon/core/schema/schema_validation.cpp:1026). SchemaValidationTest.TestLanceDataTypes asserts this rejection (src/paimon/core/schema/schema_validation_test.cpp:480).
• LanceStatsExtractor::CreateEmptyStats() has no TIME32 case either (src/paimon/format/lance/lance_stats_extractor.cpp:108). Lifting only the validation would therefore still fail when an append data file with a TIME column is finalized, with cannot create empty Lance statistics for type time32[ms]. This is the same failure VortexTimeColumnWriteInteTest.TestAppendTimeColumnFinalize guards against for Vortex.

Both sides use the same physical type, so the gap is only the C++ gating.

Solution

1. Accept TIME32 in ValidateLanceDataField(), including when nested in ARRAY and ROW. Non-millisecond units are already rejected earlier by TableSchema::Create() (Only millisecond TIME is supported).
2. Add a TIME32 case to LanceStatsExtractor::CreateEmptyStats(), following VortexStatsExtractor (src/paimon/format/vortex/vortex_stats_extractor.cpp:109), which returns empty CreateIntColumnStats.
3. No reader change is expected. The declared precision comes from the table schema (paimon.time.precision field metadata), and LanceFileBatchReader::AlignBatch() aligns each batch to the read type.

Anything else?

Suggested validation:

• Update the TestLanceDataTypes expectations so millisecond TIME, top level and nested, is accepted.
• Add a format-level round trip in src/paimon/format/lance/lance_file_format_test.cpp covering nulls and TIME nested in ARRAY and ROW.
• Add a table-level append write and finalize test, like VortexTimeColumnWriteInteTest.TestAppendTimeColumnFinalize (test/inte/write_and_read_inte_test.cpp:5984).
• Add a Java compatibility fixture with a TIME column, read from C++. The README of test/test_data/lance/append_java_compat.db/append_java_compat says "TIME32 is omitted because the C++ table path does not support it", which no longer holds. Prefer adding a new fixture over rewriting the existing one.

This needs no storage format or protocol change. TIMESTAMP_LTZ support for Lance is tracked in #476.

Are you willing to submit a PR?

• I'm willing to submit a PR!