[format] Support reading Parquet TIMESTAMP(NANOS) as timestamp(9)

[leaves12138]

Purpose

Native Parquet writers such as Arrow can encode TIMESTAMP(9) as INT64 with the TIMESTAMP(NANOS) logical annotation. This is valid Parquet, but Paimon's vectorized Parquet reader did not use the logical timestamp unit when decoding INT64 timestamp columns, so nanosecond timestamps could not be read as Paimon timestamp(9) correctly.

Changes

• Decode INT64 Parquet timestamps according to their logical time unit (MILLIS, MICROS, or NANOS).
• Convert TIMESTAMP(NANOS) schema annotations to Paimon timestamp precision 9.
• Add regression coverage for top-level timestamp(9) and array<timestamp(9)> written as Parquet INT64 TIMESTAMP(NANOS).

Tests

• mvn -pl paimon-format -DskipTests compile
• mvn -pl paimon-format -Pfast-build -Dtest=ParquetReadWriteTest,ParquetSchemaConverterTest test

[JingsongLi]

+1