Support creating tables from cudf dataframes

[ayushdg]

This PR updates the PandasInputPlugin class to also accept (optionally) cudf dataframes, to support operations like:

df = cudf.DataFrame({'a': [1,2,3]})

context.create_table("sql_table", df)

Opted to update the existing Pandas class since dd.from_pandas syntax works on cudf dataframes as well. Happy to split the implementation as well if that's preferred.

[nils-braun]

Thanks @ayushdg! Looks good to me. I do not have a strong opinion on splitting (except if there are special parameters to dd.from_pandas, but I do not see any reasons for this). However, I would opt to rename the class (and file) to something like PandasLikeInputClass, if this is fine for you.

[ayushdg]

Thanks for the suggestion @nils-braun. I've updated the class and module names to suggest that it handles pandas like dataframes.

[nils-braun]

Great! This can be merged.
One general question (not particularly on this PR): what is the best strategy to test the cudf integration into dask-sql in Github actions?

[ayushdg]

Great! This can be merged.
One general question (not particularly on this PR): what is the best strategy to test the cudf integration into dask-sql in Github actions?

GpuCI is one option that could be used to test integration with rapids libraries. It's what the dask repository uses as well (dask/community#138).
I don't think it uses GitHub actions specifically but is enabled via a webhook to GitHub. If that sounds like an option that would work for dask-sql, we could open up a broader issue for discussion where others could chime in as well

[nils-braun]

Hi @ayushdg - that is some good news. I would be perfectly fine with this solution - I will open an issue to discuss this further.