Context

`find_polars_all` (and `find_arrow_all`) crash when `parallelism="processes"` is used inside
an async context (`asyncio.to_thread`). The crash occurs deep in Python's multiprocessing
resource tracker when setting up the process pool, due to an invalid file descriptor
being passed to `fork_exec`.

Error Details

  ValueError: bad value(s) in fds_to_keep
  Stack trace (relevant frames):
  - `pymongoarrow/api.py:175` — `with multiprocessing.Pool(processes=4) as pool:`
  - `multiprocessing/util.py:524` — `_posixsubprocess.fork_exec(...)` with `passfds=(-1, 25)`

Steps to Reproduce

1. Call `find_polars_all(..., parallelism="processes")` from within `asyncio.to_thread`

Workaround

Using `promote_options="permissive"` (from related PR #390) resolves the crash.
The reporter plans to provide a minimal repro in ~2 weeks.

Additional Context

• Reported in: https://github.com/mongodb-labs/mongo-arrow/issues/391
• Related to: #390 (parallel arrow with differing schema)
• The `-1` fd appearing in `fds_to_keep` is likely the root cause — a bad fd leaking
  into the multiprocessing context when spawning from an async thread.