test: add full types boundary consistency coverage

[QuakeWang]

Purpose

Linked issue: #258

The existing Spark-provisioned full types test covered mixed Parquet, ORC, and Avro reads, but it only used ordinary non-null values and did not explicitly assert that the scan plan included all three data file formats. This left boundary values, top-level nulls, and selected nested null/empty collection cases under-covered for Java/Spark-generated data files.

This PR adds a focused consistency fixture and read test for those cases.

Brief change log

• Add full_types_boundary_table to the Spark provisioning script.
• Write Java/Spark-generated Paimon data files in Parquet, ORC, and Avro formats.
• Cover primitive min/max-style values, decimal boundaries, empty string/binary, pre-epoch date, timestamp micros, top-level nulls, nested nulls, and empty collections.
• Add a Rust integration test that:
  • verifies the scan plan includes parquet, orc, and avro data files;
  • reads the table through FileSystemCatalog;
  • checks Arrow arrays and decoded row values.

Tests

• cargo test -p paimon-integration-tests --test read_tables test_read_full_types_boundary_table -- --nocapture
• cargo test -p paimon-integration-tests --test read_tables

API and Format

Documentation

[JingsongLi]

Reviewed the new Spark fixture and boundary read coverage. The table exercises parquet/orc/avro, null values, empty collections, primitive min/max boundaries, binary/date/timestamp/decimal cases, and the Rust assertions compile cleanly against the integration test crate. I did not find blocking issues.

Verification:

• cargo test -p paimon-integration-tests test_read_full_types_boundary_table --no-run