[WIP] ivfpq: Optimize filtered batch scanning

[shyjsarah]

Summary

Optimize filtered IVF-PQ scanning by evaluating row-id filters once per relevant inverted list and sharing the resulting positions across all queries in a batch.

This is the filter-focused part split from the original combined PR. Batch-local IVF-PQ table reuse is moved to a follow-up WIP PR.

Changes

• Materialize matching row positions once per probed or loaded inverted list instead of evaluating the filter for every query.
• Evaluate in-memory filters only for lists actually probed by the current query batch.
• Use sparse position scans for selective filters while retaining dense SIMD/full-scan paths for less selective filters.
• Pass matching positions explicitly through 4-bit, transposed, batched, resident-reader, and streamed-reader scan paths.
• Preserve exact residual filtering and existing public APIs.

Testing

• cargo fmt --all -- --check
• cargo clippy -p paimon-vindex-core --all-targets -- -D warnings
• cargo test -p paimon-vindex-core --locked
• Added coverage that batch readers evaluate each loaded row once.
• Added coverage that in-memory batch search does not evaluate filters for unprobed lists.

Notes

• Public API: unchanged.
• This PR intentionally excludes batch-local list-table reuse.
• Follow-up: [WIP] ivfpq: Reuse list tables in batch search #65 adds generic filtered and unfiltered batch table reuse.