perf: offload IVF partition build to CPU pool

[BubbleCal]

before:

after:

This can improve HNSW indexing 30% faster, the other vector index types can also benefit from this but not such obvious

[codecov]

Codecov Report

❌ Patch coverage is 90.90909% with 3 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
rust/lance/src/index/vector/builder.rs	90.90%	1 Missing and 2 partials ⚠️

📢 Thoughts on this report? Let us know!

[github-actions]

Code Review

Summary: This PR offloads the IVF partition build work (filtering deleted rows, batch processing, and index building) to the dedicated CPU pool using spawn_cpu.

Assessment

The change looks correct. The spawn_cpu utility is already used elsewhere in this file (line 1349) for similar CPU-intensive operations, and the pattern is consistent.

The computation being moved (HashSet operations, filter_record_batch, and Self::build_index) is CPU-intensive and benefits from running on the dedicated CPU thread pool rather than blocking the tokio async runtime.

No blocking issues identified.