Add multi-partition Shuffle operation to cuDF Polars

[rjzamora]

Description

This PR pulls out the Shuffle logic from #17518 to simplify the review process.

The goal is to establish the shuffle groundwork for multi-partition Join and Sort operations.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[rjzamora]

@madsbk - Just curious: When we integrate with rapidsmp, can we be able to simply call pack on the elements of this mapping and hand it off to the shuffle service?

[madsbk]

Yes, I think we can use partition_and_pack()

[rjzamora]

Okay cool - That sounds good for a hash-based shuffle, but we will not want hash partitioning for sorting. Rather, we will want to pass in a pylibcudf column containing the final partition for each row. Do we need to add that utility in rapidsmp?

[madsbk]

Yes, we would need an util function that takes a table and split+pack it into a dict[PartID, PackedColumns] based on the values in a column.

[rjzamora]

cc @wence- - It may make sense to get this in before Join or Groupby support. The Join logic largely depends on shuffling, and GroupBy may take a bit longer to clean up. (I can also push on Sort once the shuffling foundation is in place).