GeoPandas: implement GeoDataFrame.to_feather

Part of #2230.

Implement `GeoDataFrame.to_feather`, which currently raises `NotImplementedError`.

Keep the geopandas signature (`path`, `index`, `compression`, `schema_version`, `**kwargs`), which the existing stub already matches, and pass extra keywords through to `pyarrow.feather.write_feather`.

Feather is a single file format and Spark has no writer for it, so the proposed approach is to collect the frame to the driver and let GeoPandas write it. `GeoDataFrame.to_arrow` also delegates to `to_geopandas()`, for a different reason: it returns an in-memory GeoPandas object that has no Sedona equivalent. I have a working implementation and tests locally, and I would appreciate feedback on whether driver side collection is acceptable for a file writer before I open the PR.

Document the driver collection cost and the index behavior. Because Sedona converts `RangeIndex` into a general `Index`, `index=None` writes the index, and reading the file back gives an `Index` where geopandas gives a `RangeIndex`. `to_arrow` documents a related caveat, but its wording about the index appearing as a column does not carry over to Feather, where pandas restores it as the index.

Tests should cover `index=False`, `index=True`, `index=None` and keyword pass-through, reading back with `geopandas.read_feather`. Sedona has no `read_feather` of its own, so the Sedona half of the round trip goes through `GeoDataFrame(geopandas.read_feather(path))`.

cc @petern48

Start at the GeoDataFrame.to_feather stub and compare its existing signature with GeoDataFrame.to_arrow; inspect how the GeoPandas writer handles path, index, compression, schema_version, and extra keywords. Add coverage for index=False, index=True, index=None, keyword pass-through, and round trips through geopandas.read_feather and GeoDataFrame, including the documented driver-collection cost and index behavior.

Written by the indexing model from the issue text.