[Python] `pyarrow.write_feather` can't be used in `atexit` contexts to write a `pandas.DataFrame`

### Describe the bug, including details regarding any error messages, version, and platform.

When `pyarrow.write_feather()` is given a `pandas.DataFrame`, `write_feather()` unconditionally calls `Table.from_pandas()` with the default `nthreads=None` argument. This is then passed to `pandas_compat.dataframe_to_arrays()`, allowing it to heuristically use a `concurrent.futures.ThreadPoolExecuter` to convert columns. This causes a runtime error when `write_feather` is used in an `atexit` (or `weakref.finalize`) context on exit of the interpreter:
```
RuntimeError: cannot schedule new futures after interpreter shutdown
```
This scenario can be avoided by adding a `use_threads` parameter to `write_feather` that can be used to force serial operation.

### Component(s)

Python