fix: join Ray pools after closing

[hfutatzhanghb]

Summary

• Call pool.join() after pool.close() in the Ray Pool cleanup paths for distributed indexing, add_columns, and compaction.
• Add a unit test covering the shared index Pool helper cleanup order.

Why

ray.util.multiprocessing.Pool follows the multiprocessing-style lifecycle: close() prevents new work and lets outstanding actor work finish, but it does not wait for the Pool actors to exit. join() is the call that waits for actors in a closed Pool to stop.

These code paths already wait for map_async(...).get() before returning results, so this does not change result collection semantics. It makes worker actor cleanup deterministic after successful and failed runs, reducing the chance of transient Ray actor/resource leftovers across repeated operations or tests.

Validation

• python -m py_compile lance_ray/index.py lance_ray/io.py lance_ray/compaction.py tests/test_vector_index_options.py
• python -m pytest tests/test_vector_index_options.py
• uv run --no-sync ruff check lance_ray/index.py tests/test_vector_index_options.py
• git diff --check

[hfutatzhanghb]

@chenghao-guo Hi, could you help reviewing this PR when have free time? Thanks very much!!!