perf: don't concat the batches for writing posting lists

[BubbleCal]

the file writer has an internal buffer so we don't need to concat the batches, this slightly improves the indexing perf (6%) and reduces the memory footprint (-4%)

[github-actions]

Code Review

This PR removes the custom batching logic in write_posting_lists and relies on the file writer's internal buffer instead. The change looks correct and well-tested.

Summary

The PR simplifies the posting list writing code by removing the LANCE_FTS_FLUSH_SIZE environment variable and the associated manual batching logic. Since FileWriter already has internal buffering (8MiB per column by default via data_cache_bytes), this external batching was redundant.

Observations

1. Correctness: The change is sound. The file writer's write_batch method already buffers data internally before flushing pages, so the manual concat_batches + size-based flush was unnecessary overhead.

2. Test Coverage: Good addition of test_write_posting_lists_writes_each_batch with a mock CountingStore to verify the new behavior writes each batch individually.

No blocking issues found.

[codecov]

Codecov Report

❌ Patch coverage is 59.18367% with 20 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
rust/lance-index/src/scalar/inverted/builder.rs	59.18%	19 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!