[python] Expand Native update input parity coverage

[JingsongLi]

Purpose

Expand parity coverage for existing public PyPaimon update inputs that need companion Native core fixes: MAP entries, encoded Union values, fixed-size binary strings, Float16 strings, and exact/Java-compatible decimal conversion.

Depends on apache/paimon-rust#1149. Native CI continues to build Rust main. This PR stays draft until the core change is merged.

Brief change log

• Cover batch/stream row-ID updates, incremental updates, predicate array/callable assignments, matched upserts and MERGE source-column assignments, only where Python already accepts the input.
• Exercise MAP-to-ARRAY/ROW input adaptation at the root and inside ROW/ARRAY/MAP containers, renamed fields, slices, independent chunks, NULLs and empty containers.
• Check fixed-size binary UTF-8/empty strings, invalid UTF-8 rejection, exact Float16 text, and Decimal128/256 floating rounding.
• Cover Decimal32/64/128/256 exact rescaling and zero/NULL across extreme physical scales without relaxing overflow or fractional-loss checks.
• Exercise dense/sparse Union primitive/ARRAY/ROW children, nonstandard IDs, inactive invalid values, dictionary NULL indices versus active NULLs, NULL parents, missing ROW fields, root MAP constructor scope, and unaligned foreign buffers.
• Assert safe predicate/MERGE assignments still reject unsupported Union casts, append checks remain strict, and rejected updates do not commit.
• Verify Python/Native reads, stable row IDs and untouched values. Fail on unintended Python update/read fallback when Native is selected.

Only tests are added. Python production code and public APIs remain unchanged; matching, normalization and conversion stay in Rust core, with private layout handling at its Arrow C Data boundary.

Review follow-up

• Fix Decimal256 source precision 39..76 and coefficients wider than i128 when assigning FLOAT/DOUBLE. Use complete decimal coefficients while retaining Java compact/wide rules for valid logical DECIMAL inputs.
• Normalize ordinary nested STRUCT siblings and active Union children along with sparse Union slices; avoid applying nested offsets twice.
• Defer NULL/non-nullable-child validation until parent and child coordinates are both normalized, including masked STRUCT and FixedSizeList values.
• Add 264 public-API regression cases. All three findings were reproduced before the fix; subsequent independent review found no remaining actionable issues. Java 8 BigDecimal results were checked directly.

Tests

• Old Native failures reproduced while the corresponding Python API reference cases passed.
• All 210 MAP/container reference cases and 795 scalar/encoded reference cases passed before Native verification.
• Built the companion extension before Native tests; no build/install overlapped Native tests.
• New and existing nested update, selected-value upsert, match-key, Blob, MERGE and row-shape regressions: 2669 passed, 2 skipped, with Native plan/read/write/commit enabled. Counters: 3243 plans, 4036 reads, 1451 writes, 147 commits.
• Flake8 using dev/cfg.ini and diff checks passed.

[leaves12138]

LGTM