feat(index): apply shared FRI remapping to scalar and vector queries

[LuQQiu]

Stack reordered: the standalone format proposal is now #9136 (replacing #9065), followed by #9106 → #9064 → #9067 → #9068 → #9107 in native stack #9137. This PR’s diff is byte-identical to its pre-reorder diff. All branches include main 31d78d170; no history was rewritten. Formatting checks pass; Clippy and tests remain blocked by the unavailable object_store_opendal 0.60.1 dependency.

Connect scalar and vector index loading to the common FRI reader from #9068. Mapping algorithms and lineage traversal remain in earlier PRs; this PR adapts index consumers and enables derived coverage only for supported consumers.

• Share asynchronous batch remapping across B-tree, bitmap, zonemap, bloom, FTS, ngram, label-list, RTree, and vector partition loading. Preserve vector layouts and deletion handling without a partition-sized per-row lookup table.
• Keep index-version-0 FRI on the existing synchronous reader/writer path: scalar cold loads call the original plugin API, and vector storage receives the original synchronous remapper. Only external mappings enter the new asynchronous loading branch. Unrelated FM/custom plugins retain legacy loading; unsupported affected consumers scan.
• Retain legacy translated-index cache isolation and support the existing vector prewarm entry point. Immutable ledger/mapping reuse is in feat(index): compose FRI mapping readers with planner safety boundaries #9068. New label-block caching, row-map prewarm optimizations, and the core FRI writer are later work.

Validation:

• A two-source/two-destination end-to-end matrix verifies every contributing segment is queried, and that missing segments, unsupported index versions, or unsupported async consumers force scan fallback without missing or duplicated rows.
• Vector fragment-restricted search falls back to scanning when explicit segment selection omits a contributor; segment-only partial-search semantics remain unchanged. Regressions cover default, complete, and partial segment selection, including a single destination with a prefilter.
• Queries prefer direct destination coverage and combine indexed fragments with scans for uncovered fragments. When branches reconverge, each segment excludes paths entering fragments supplied by other selected segments, preventing duplicate vector results.
• 79 common-reader and index-consumer tests pass, including scalar/vector index-version-0 lifecycles, released legacy fixtures, inline/external history, eager/deferred compaction, append/delete/checkout, RTree, unrelated FM, unknown mappings, and maintenance boundaries.
• Default-budget local run: cargo test -p lance frag_reuse_reader --lib -- --test-threads=1.
• Run serially with the default memory budget. The default 150 MB process-shared DataFusion pool can exhaust sort reservations with concurrent tests; CI uses nextest process isolation.
• Shared remapper unit tests pass (batch bounds, deletion/layout preservation, legacy delegation).
• cargo fmt --all and workspace cargo clippy --all --tests --benches -- -D warnings pass.

No format documentation changes. Draft, last layer of the six-PR reader stack.

Latest main synchronization (2026-09-10): merged d7b031f9c through the stack without rewriting history. cargo fmt --all and whitespace checks pass. Workspace Clippy could not run: main now requires object_store_opendal 0.60.1, but the crates.io index currently resolves only up to 0.60.0. Tests were not rerun after this synchronization because dependency resolution is blocked. Earlier validation above predates this merge.

[lance-gatekeeper]

✅ Gate recommendation: approve.

The synchronized base and head only move the tagged FRI capability flag to its non-conflicting bit; this PR retains the same stable patch ID as the reviewed revision. The shared batched remapping boundary still keeps v0 on the legacy path, scopes tagged caches by snapshot, and falls back to scans for unsupported or incomplete coverage; focused scalar, vector, lifecycle, and prewarm coverage remains successful.