[FEA]: Expose CUDA streams in cuda.parallel APIs

[shwina]

Is this a duplicate?

• I confirmed there appear to be no duplicate issues for this request and that I agree to the Code of Conduct

Area

cuda.parallel (Python)

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

Consider the reduce algorithm. The underlying C library exposes a stream argument. On the Python side however, we don't accept a corresponding stream argument, and always pass None (nullptr, or "default stream") to the C API. Thus, Python users are unable to take advantage of concurrency via CUDA streams.

Describe the solution you'd like

We should change the Python API to accept a stream argument, and pass that to the underlying C library.

What should the argument type be?

There are a few options here, that I'm ordering from least preferred to most (in my opinion):

1. Accept an int (least preferred)
2. Accept a concrete numba stream object.
3. Accept a concrete cuda.core.Stream object.
4. Accept an object with the __cuda_stream__ protocol - this would automatically include (3) without an explicit dependency on cuda.core.

Describe alternatives you've considered

No response

Additional context

No response