[python] Enable supported native Parquet writer capabilities

[JingsongLi]

Purpose

Enable PyPaimon to use the Parquet capabilities already implemented in Rust core and remove obsolete native writer/read gates.

Depends on apache/paimon-rust#970. This PR is draft until that change reaches Rust main. Native CI now builds apache/paimon-rust@main and continues to print the resolved commit for provenance; it does not pin the companion topic branch.

• Enable native batch/stream writers for ordinary Parquet data-evolution tables, first-row/partial-update/aggregation merge engines, primary-key value statistics, batch-boundary row rolling, and compatible nested/fixed-size binary Arrow input.
• Keep fallback for unsupported MAP shared-shredding and data-evolution row sidecars, along with existing format/Blob/bucket constraints.
• Defer Python-only merge capability validation until a read actually falls back to Python. Invalid configurations still fail eagerly; unsupported Python merge options cannot bypass validation through raw splits, streaming readers, iterator APIs or parallel dispatch.
• Share canonical partition-path resolution among append/PK, data-evolution and chunk-shuffle planners. This lets Python readers locate Java/Rust escaped partition directories while retaining legacy Python paths and explicit external paths.
• Keep implementation-specific assertions on their Python lane and make Ray fixture batches explicit so batch-boundary rolling still yields deterministic file groups.

Tests

Against the locally rebuilt companion Rust tree, with native plan/read/write/update/commit all enabled:

• Affected Native, data-evolution, merge-engine, Parquet page-index, Variant, nested-input, Blob, sequence and Ray suites: 1,686 passed plus 186 subtests. The run identified an unpartitioned path-helper regression, fixed and rechecked below. The remaining three cases require the locally unavailable optional Vortex package.
• Final path-helper/group-stats/new native-capability regression run: 47 passed plus 17 subtests.
• Scanner, deletion-vector-path and sequence regression run: 218 passed.
• Rust Python write binding tests: 29 passed.
• Flake8 for every changed Python file and git diff --check passed.

The new capability tests assert the native writer is selected and native reads do not fall back; check results through both readers where their capabilities overlap; inspect primary-key value statistics; cover sliced nested/null input, repeated commits, sequence-group aggregation, delete filtering, partitioned delete/history reads and chunk shuffle.

GitHub Native CI must be rerun after apache/paimon-rust#970 merges because this PR intentionally follows Rust main.

[leaves12138]

Re-reviewed 4cbe30d against paimon-rust 5c87ec53e8bbaa415047e9a5a2beba27fe1b0acf. No remaining blocking issues found in the reviewed changes.

The full-data-types test now deterministically covers explicit full and none statistics modes, with separate tables and native writer coverage preserved.

Local validation on this revision:

• Python reader suite: 31 passed.
• Native planning, reading, writing, updating, and committing regression suites: 587 passed.
• Additional Ray, Variant, nested schema evolution, MAP/Blob fallback, and reader regression suites: 451 passed in this re-review.
• Flake8 and whitespace checks passed.

GitHub CI is still running. This approval reflects code review and local validation; please let the remaining CI checks pass before merging.