fix: align format options and listagg semantics with Java

[lxy-9602]

Purpose

This change aligns several format-related behaviors with the latest Java Paimon implementation:

• Keep legacy manifest formats readable, but allow only Avro for manifest writes.
  • manifest.format is retained as a deprecated compatibility option.
  • Writing with a non-Avro manifest format now returns an explicit error.
  • Read paths continue to use the configured legacy format.
• Convert the parquet/append_09 test fixture manifests from ORC to Avro while preserving the existing manifest filenames and snapshots.
• Rename the Parquet dictionary option from parquet.enable-dictionary to parquet.enable.dictionary, matching Java Paimon.
• Align listagg blank-string handling with Java:
  • Add StringUtils::IsBlank with Java Character.isWhitespace semantics.
  • Ignore empty, whitespace-only, and supported Unicode blank inputs.
  • Prevent blank accumulators from producing a leading delimiter.
  • Ignore blank tokens when distinct is enabled.

Tests

Added or updated unit tests covering:

• read and write access for Avro and legacy manifest formats
• rejection of non-Avro manifest writes
• disabling Parquet dictionary encoding with parquet.enable.dictionary
• ASCII and Unicode behavior of StringUtils::IsBlank
• listagg handling of blank inputs, accumulators, and distinct tokens
• commit tests using the Avro-converted parquet/test_data fixture

API and Format

Behavioral and format-related changes:

• Avro is now the only writable manifest format.
• Existing non-Avro manifests remain readable.
• The Parquet dictionary option is renamed to parquet.enable.dictionary.

Documentation

Generative AI tooling

Generated-by: Codex (GPT-5)