feat(external): skip Parquet RowGroups via min/max statistics filtering

[flypiggyyoyoyo]

What type of PR is this?

• API-change
• BUG
• Improvement
• Documentation
• Feature
• Test and CI
• Code Refactoring

Which issue(s) this PR fixes:

issue #23703

What this PR does / why we need it:

ParquetReader currently decodes every RowGroup regardless of WHERE conditions. This PR adds RowGroup-level predicate pushdown: before reading each RowGroup, we evaluate the filter expression against ColumnChunk min/max statistics using EvaluateFilterByZoneMap. RowGroups that cannot match are skipped entirely, avoiding unnecessary page decoding.

Key changes:

• New file parquet_filter.go: bridges Parquet ColumnChunk statistics to MO's ZoneMap evaluation via parquetColumnMetaFetcher adapter
• ParquetHandler.prepare(): initializes RowGroup list and filter column mapping; defers page opening when filtering is possible
• ParquetHandler.getData(): routes to getDataByRowGroup() when canFilter=true, which skips non-matching RowGroups before opening their pages
• Supported filter types: bool, int8–64, uint8–64, float32/64, date, timestamp/datetime. String types fall back to no filtering (TODO)
• Unit tests (22) and e2e tests (multi-RowGroup parquet file with various WHERE patterns)