fix(index): reject user-specified index_uuid for segmented builds

[zhangyue19921010]

closes: #7032

What & why

The segmented-index architecture generates segment UUIDs internally — each segment is written to its own indices/<uuid>/. Accepting a user-specified index_uuid there is unsafe: it is used verbatim as the directory, so concurrent distributed workers (or a collision with an already-committed index) can silently overwrite each other's segment files. Once an index type is built on this architecture, a user-specified UUID has no valid use.

This adds a single guard in the Rust core (CreateIndexBuilder::execute_uncommitted, the one chokepoint all language bindings funnel through) that rejects index_uuid for any index built via the segmented-index path. It is keyed on the existing uses_segment_commit_path predicate, so it automatically covers each index type as it migrates to the architecture — no per-type wiring.

Behavior changes

1. Indexes built on the segmented-index architecture no longer accept index_uuid.
   They reject it with:

   index_uuid is no longer accepted for <type> indexes; segment UUIDs are generated by Lance and returned in the index metadata.

   This applies to both full and fragment-scoped/distributed builds. Callers omit it and read the generated UUID from the returned index metadata.

2. (Bug fix, side effect) The segment-commit path now actually runs for these indexes.
   It was previously dead code — the predicate compared the user-facing index name against the internal __lance_vector_index kind id and was always false — so builds fell through to the legacy single-index commit. Output is equivalent and all existing tests pass.

Scope / impact

• Affected now: index types currently on the segmented-index architecture (vector). Passing index_uuid via Python (create_index / create_index_uncommitted), Java (IndexOptions.withIndexUUID), or Rust (CreateIndexBuilder::index_uuid) now errors.
• Not yet affected: index types still on the legacy distributed merge_index_metadata flow (scalar: BTREE / BITMAP / INVERTED / …) keep accepting a shared index_uuid. They are picked up by the same guard automatically once migrated — no API change needed then.
• No public API signature removed — index_uuid parameters stay; the rule is enforced centrally in the core.
• Benchmark: distributed_vector_build was rebuilt around the segmented merge API (merge_existing_index_segments); it previously drove the now-removed shared-UUID merge_index_metadata path.

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!