[python] Enable Parquet page indexes when supported

[XiaoHongbo-Hope]

Purpose

Enable Parquet page indexes by default when supported (PyArrow >= 13), preserving older-version compatibility. parquet.write-page-index.enabled=false disables them; explicit true rejects unsupported versions before creating files.

Tests

286 local tests passed. Verified all three states, vector files, PyArrow 6/12 compatibility, and Java Parquet/Rust readback. Lint, license checks, and git diff --check passed.

[JingsongLi]

I think the default behavior should align with the Java writer: write the Parquet Page Index whenever the installed PyArrow version supports it, instead of defaulting to false.

Paimon Java, through parquet-java, already writes ColumnIndex and OffsetIndex by default. Keeping Python disabled by default means that files written to the same table have different metadata depending on the writer implementation.

A plain default_value(True) would break the PyArrow 6/7 versions that PyPaimon still supports, so I suggest capability-aware, three-state semantics:

• unset: enable automatically on PyArrow >= 13; omit the argument on older versions;
• explicit true: enable it, and fail before creating output when PyArrow < 13;
• explicit false: always disable it.

This could be implemented by making the option have no configured default and resolving the default in create_parquet_writer_options based on the installed PyArrow version. The documentation could describe the default as: "enabled when supported by the installed PyArrow version (PyArrow >= 13); otherwise disabled."

The tests should then expect setting=None to produce Page Indexes on PyArrow >= 13, while a mocked PyArrow 12 should still write successfully without passing write_page_index.

The option name itself looks reasonable to me: "Page Index" is the Parquet specification term, and write-page-index is consistent with PyArrow write_page_index and Impala PARQUET_WRITE_PAGE_INDEX.

[JingsongLi]

+1