[WIP] Leverage fused RapidsMPF Join operation

[rjzamora]

Description

• Blocked by Add Dask integration for a join operation rapidsmpf#517
• First step on the path to better join behavior in multi-GPU Polars
  • Next step is "fused" broadcast join support using the same RapidsMPF API

Checklist

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

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[TomAugspurger]

These types need to be updated to match https://github.com/rapidsai/rapidsmpf/pull/517/files#diff-f053a4fe09712b51b39f32074476d744af1a34774cae7b50a89a49c5b2d7b551R291 I think (cudf.DataFrame | Callable[[int], cudf.DataFrame])

[rjzamora]

Good catch - This needs to be Callable[[int], DataFrame] now.
While prototyping the bcast join, I realized the backend-specific code is much simpler if we are always fetching the partition chunks with a callable.