Improve support for Parquet in abstract files sources (e.g. S3)

[Phlair]

Parquet support in this PR. Currently we're not actively supporting partitioned parquet datasets, but this is quite a common use-case.

The way abstract files source works iterating through file by file makes reading a partitioned parquet dataset hard. See how PyArrow can handle this here and here.

At the moment I think the connector would work sort of, but it could have quite poor performance and more importantly miss out on the columns that the parquet dataset is being partitioned on (untested but I think that would be the case).

A couple of enhancements that would be good here:

• Make adjustments to abstractions in abstract files source to allow a format parser to take in multiple file paths for more optimised processing
• Extend parquet reader using above changes to handle Parquet datasets better (see links above to PyArrow docs)
• Could even allow custom cursor fielding in the case where Parquet is partitioned on date-like updated_at values, further improving performance potential.

[artem1205]

Done in #25937