feat(python): Write PyArrow types directly to Vortex

[a10y]

Fixes: #4160

Extend PyVortex write API to accept PyArrow Tables and RecordBatchReaders
directly, enabling streaming writes without loading entire datasets into memory.

Key changes:

• Support PyArrow Table and RecordBatchReader objects via ArrowArrayStreamReader
• Stream Arrow RecordBatches directly to Vortex ArrayIterator

[coveralls]

coverage: 84.935% (+0.06%) from 84.877%
when pulling 04e9b0a on claude/issue-4160-20250807-2038
into ac773cb on develop.

[kylebarron]

👋 Just saw this and have a small suggestion: this current implementation imports pyarrow explicitly and thus will only work if pyarrow itself is installed in the user's environment. But since Arrow is a stable format, you really don't need pyarrow; this is why the Arrow PyCapsule Interface was created. Any Python object that exposes an __arrow_c_stream__ method can be called to get a PyCapsule with an Arrow C Stream pointer.

There are a bunch of libraries that already support this, so if you implemented support for the PyCapsule interface then you could take in data from Polars/DuckDB/DataFusion etc even if pyarrow isn't installed in the environment.

You can use pyo3-arrow (disclaimer: my library) which has some benefits over the arrow-rs Python integration. Or, I think the arrow-rs integration should use it automatically, and you shouldn't need to import pyarrow explicitly.

(I could make an issue to discuss this if you prefer)