feat: support `read_parquet`/`read_csv` for most backends

[jcrist]

trafficstars

Currently we only support read_parquet for backends that have native support (like duckdb). In contrast, we support to_parquet for all backends, falling back to a common pyarrow implementation if a backend doesn't natively support it.

To provide more uniform feature coverage, we could write an equivalent common pyarrow (or other) implementation of our IO input methods (read_parquet/read_csv/...) that backends like postgres could fall back on.

[jcrist]

If duckdb is installed, there's also the option of using duckdb for faster support for parquet/csv table loading. I think that could be done in a 2nd pass, see #8110, but a pass using pyarrow first to cover all backends makes sense to me as a good first step.

[jitingxu1]

HI @jcrist I could take this one. Will do it one by one.

Plan to add read_parquet and read_csv. How about read_json and read_delta?

[csubhodeep]

Any updates on this one?