perf(fts): push prefilter through scalar index on flat FTS

[LuQQiu]

Summary

• Flat FTS (no inverted index on the text column) used scan_fragments + a manual LanceFilterExec to apply the prefilter, bypassing any scalar index on the filter column.
• Route the scan through filtered_read instead — same pattern as the brute-force KNN path at scanner.rs:3839-3848 — so the pushable part of the prefilter is evaluated inside FilteredReadExec (using a scalar index when one exists) and only the unpushable refine_expr is reapplied on top.
• Requires prefilter(true) on the scanner. The postfilter branch passes an empty ExprFilterPlan::default() down the FTS path and is unaffected.

Plan shape

Before (with WHERE id = 1 and a BTree on id, no FTS index on text):

FlatMatchQueryExec
  └── LanceFilterExec(id = 1)            # post-scan filter, BTree unused
        └── LanceScan(columns=[text, id], with_row_id)

After:

FlatMatchQueryExec
  └── FilteredReadExec                   # full_filter=id = Int32(1), BTree used
        columns=[text], with_row_id
        # refine_expr (if any) reapplied as LanceFilterExec on top

Test plan

• New test_fts_without_index_uses_scalar_index_for_prefilter in dataset_index.rs: BTree on id, no FTS index, flat FTS + id = 1 prefilter with .prefilter(true). Asserts via analyze_plan that LanceRead shows full_filter=id = Int32(1), no LanceScan: in the plan, and the result set is correct (2 rows).
• Full FTS test suite passes (123 tests).
• cargo fmt --all
• cargo clippy --all --tests --benches -- -D warnings

[LuQQiu]

Duplicate; reopening against lancedb/lance via API.

[codecov]

Codecov Report

❌ Patch coverage is 70.00000% with 6 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
rust/lance/src/dataset/scanner.rs	70.00%	2 Missing and 4 partials ⚠️

📢 Thoughts on this report? Let us know!