Standardize on Arrow RecordBatchReader in function parameters and return types

[kylebarron]

This should be a pretty easy change that I want to make before the next release and comes with a couple benefits.

• Allows you to access the schema of the stream without having to "peek" at the stream manually, like we do currently

  stac-geoparquet/stac_geoparquet/arrow/_to_parquet.py

  Line 51 in 768df20

  first_batch = next(batches_iter)

• Allows direct zero-copy interop with compiled code. E.g. @gadomski 's stac-rs arrow support would be able to directly use the result of stac_geoparquet.arrow.parse_stac_items_to_arrow without having to call back into Python to iterate over the loop. It also means you'd be able to pass the result of parse_stac_items_to_arrow directly into, say, pyogrio.write_arrow to write via GDAL. This is all thanks to the Arrow PyCapsuleInterface