[FEA]: Update cudf-polars join functions to use CUDA streams #20240
Description
The initial implementation of adding CUDA streams to cudf-polars (#20154) doesn't maximize the concurrency. In particular, see
- https://github.com/rapidsai/cudf/pull/20154/files#r2426527192
- [cudf-polars] CUDA stream #20154 (comment)
There are some subtleties around the stream ordering of some sequences of operations that mix values that can happen independently on multiple streams and values that are on a joined stream.
This issue will be closed once we've maximized the available concurrecny.