fix(storage): preserve Parquet schema annotations

[youngsofun]

I hereby agree to the terms of the CLA available at: https://docs.databend.com/dev/policies/cla/

Summary

Parquet table plans serialized their inferred schema by printing and reparsing the Parquet message type. That round-trip could normalize a legacy Decimal annotation from converted_type=DECIMAL, logical_type=None into an explicit Decimal logical type. Row-group readers then rejected the original file because its schema no longer exactly matched the plan.

This PR:

• serializes the Parquet schema through its binary metadata representation so legacy annotations are preserved exactly
• reports the first concrete Parquet schema difference, including the actual file and schema-inference source paths, while treating producer-specific root message names and field IDs as compatible for stage reads
• reports field-count mismatches with both file-only and inferred-only field names; each list shows up to five names and its remaining count instead of dumping the full schema
• adds direct schema-bytes and full plan serde round-trip tests that verify a legacy Decimal annotation remains unchanged
• adds logic and unit coverage for compatible field IDs, readable first-difference diagnostics, field-list direction, and truncation

Mixed-version query nodes retain the existing Arrow-derived fallback when they cannot decode the other plan format. That fallback is best-effort and may normalize legacy annotations; subsequent schema mismatches now identify this condition and suggest checking query-node versions. Rolling-upgrade correctness for this case is outside this PR's scope.

Tests

• Unit Test
• Logic Test
• Benchmark Test
• No Test - Explain why

Validated with focused catalog serialization tests, Parquet storage schema-difference tests, clippy for the affected crates, formatting, shell syntax, and diff checks. The schema-difference logic test reuses existing Parquet fixtures to cover file-only/inferred-only direction, complete short lists, and truncated wide-schema lists.

This regression does not keep an integration fixture. A tiny single-file stage read does not guarantee that the scan plan is serialized to another query node, so it can pass with the old implementation. Using several tiny files would still depend on scheduler placement rather than deterministically exercising the faulty round-trip. The direct schema-bytes round-trip and full ParquetTableInfo serde tests instead reproduce the normalization at its source and assert that logical_type=None is preserved.

Type of change

• Bug Fix (non-breaking change which fixes an issue)
• New Feature (non-breaking change which adds functionality)
• Breaking Change (fix or feature that could cause existing functionality not to work as expected)
• Documentation Update
• Refactoring
• Performance Improvement
• Other (please describe):

AI assistance

• AI usage: An AI coding agent investigated the Parquet schema round-trip, drafted the storage fix and regression tests, and assisted with validation and the PR summary
• Responsible human: @youngsofun
• The responsible human has read every line of this diff and can explain each change

---

This change is