PyVortex DuckDB to use native extension

[gatesn]

The PyVortex <-> DuckDB integration should (somehow) use the native DuckDB extension, instead of going via Arrow datasets.

I've opened this as a discussion because I haven't really thought about how to do this!

[paultiq]

If you're just talking about Vortex datasets registered for replacement scans in DuckDB-Python, then:

DuckDB-Python handles all Arrow datasets the same: wrapping them in a Scanner and pushing the expressions down.

One way to avoid the expression issues from #5801 is to special case Vortex datasets: adding a separate case for Vortex datasets that uses Vortex expressions... possibly reusing the Vortex extension.

The relevant code (that handles Datasets and creates the Scanner):

https://github.com/duckdb/duckdb-python/blob/694922b261e5d5aa546961c67ce5f88cf10a3782/src/duckdb_py/arrow/arrow_array_stream.cpp#L105

There's other places to inject a special case, such as at registration time: all involving DuckDB-Python changes.