[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.