[DOC] Document cloud/cluster engine recommendations for cudf-polars

[Matt711]

We have benchmark evidence pointing to configuration choices for cloud deployments, for example pinned memory defaults, but none of that guidance is written down anywhere. We should document recommended engine and configuration choices for running cudf-polars on cloud infrastructure, including benchmarks.

[Matt711]

xref #23836

[vyasr]

I think this goes hand in hand with #22892 (comment). Once we have the benchmarks in a package, we can start documenting optimal benchmark configurations in that package, and over time we can hopefully propagate that backward into better defaults in the main cudf-polars package itself. To the extent that manual tuning remains necessary we can then add suitable documentation as requested in this issue.