[python] Expand Native upserts, Dictionary MERGE and Parquet vector updates

[JingsongLi]

Purpose

Expand Native read/write/update execution for capabilities PyPaimon already supports, keeping matching and conversion in Rust core.

• Keep Arrow upserts on Native when surviving matched payloads need existing row-ID conversions, or when ignored/discarded values do not satisfy the append schema. Newly appended rows retain ordinary write validation.
• Qualify logical upsert keys across exact integer widths, decimal scales and temporal units, Dictionary encodings, and NULL-typed input. Rust validates values, overflow and precision; original input keys remain subject to their action's write rules.
• Enable Dictionary MERGE keys, including Pandas categorical sources and partition keys. Rust decodes keys and applies SQL NULL/cardinality semantics; Python adds no matching implementation.
• Enable row-ID, incremental, predicate, upsert, MERGE and shard updates for Parquet vector sidecars, including coexistence with BLOB and Paimon table MERGE sources. Reuse existing core writers and dedicated-file metadata.
• Preserve split order when the Python shard updater uses Native reading. Parallel completion order otherwise associates update values with other row IDs. A regression reads actual files in reversed group completion order to reproduce the problem deterministically.

Matched values reuse the existing core row-ID conversions, including Binary/LargeBinary; ordinary append validation keeps its existing rules. No new dependency, public API, metadata path, or storage format.

Depends on apache/paimon-rust#1147. Native CI continues to build Rust main, so this PR remains a draft until the core changes are merged.

Tests

Built the Native extension from the companion Rust branch before starting Python tests, then ran the selected-value, key normalization, Dictionary MERGE, Parquet vector, existing upsert layout/row-shape, Blob row/update, nested update, general Native update/MERGE and shard update suites with PYPAIMON_TEST_NATIVE_PLAN=1, PYPAIMON_TEST_NATIVE_READ=1, PYPAIMON_TEST_NATIVE_WRITE=1 and PYPAIMON_TEST_NATIVE_COMMIT=1.

Result: 1322 passed, 2 skipped. Native counters: 1887 plans, 2236 reads, 809 writes, 257 commits. Flake8 with the repository config and git diff --check passed.

New tests compare Python/Native batch and stream execution, fail explicitly on unintended Python update fallbacks, and verify values, unchanged columns, partitions, stable row IDs and dedicated vector/BLOB metadata. Key coverage includes exact conversions, overflow/precision rejection, duplicate last-write-wins, target fan-out, NULL indices and NULL dictionary values, partition dictionaries, Dictionary MERGE cardinality and unconditional DELETE, plus original append type validation. 32 positive Python/Native batch and stream cases verify Binary/LargeBinary conversions for arbitrary bytes, NULLs, nested MAP/ROW values and constructor-shaped inputs, with stable row IDs.

Follow-up review fixes

• Remove the upsert-specific Binary/LargeBinary restriction and its recursive policy plumbing in the companion core PR. The Python bridge adds no conversion policy; matched payloads reuse the established row-ID conversions.
• Preserve Blob map transport and Java key serialization independently of value conversion: Dictionary keys and Dictionary payloads, exact integer widths, DECIMAL HALF_UP and TIME millisecond normalization. Normalization-induced duplicate keys still fail, and append types remain strict.
• Handle exact decimal zero/NULL keys across extreme scale differences without Arrow power-table errors or i8 scale arithmetic overflow. Nonzero overflow and precision loss remain errors.
• Add regression cases for those failures, NULL containers, mixed Blob map transport and unchanged row identities. Several independent review passes found no remaining confirmed issue after repairs.

[leaves12138]

LGTM