Is your feature request related to a problem? Please describe.

As part of cudf-polars' work to use CUDA streams, we need a way to join multiple CUDA streams (in, e.g., a join operation) to avoid race conditions.

Describe the solution you'd like

cudf (C++) offers a join_streams method:

. A python API in pylibcudf that wraps that should work for our use cases.

Describe alternatives you've considered

Reimplementing the logic in cudf-polars.

Additional context