[FEA] Expose alignment-explicit allocation routines through cython interface

[wence-]

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

With the move to rmm::device_async_resource_ref, the most general form of the allocation functions are the allocate_async ones that accept size_t bytes, size_t alignment, cuda_stream_view stream. We should consider how to expose and/or use these from cython.

[bdice]

@wence- Is this something you'd like to work on for 25.02? We have it on our 25.02 roadmap.

[wence-]

I might have time, but I am not sure it is high priority. I would be happy to let it slip a bit.

[bdice]

Same here. I'll move to 25.04 pre-emptively, as we have plenty on our plates for 25.02.