[cuda.compute]: Add support for accepting DLPack arrays

[NaderAlAwar]

In cuda.compute, we currently require that input arrays implement __cuda_array_interface__. This allows us to support Python arrays from different libraries in cuda.compute since it is a standard interface to extract array metadata.

DLPack is a more recent standard supported by a larger number of array libraries. We should allow users to pass in DLPack arrays instead of only __cuda_array_interface__.

cuda.core includes StridedMemoryView which extracts array metadata from arrays that implement both __cuda_array_interface__ and DLPack. We should use StridedMemoryView in cuda.compute.

cc @shwina

[catswe]

Hi. I think I can take this on if no one is working on it yet

[shwina]

Hi @catswe - thank you for your interest in contributing! I don't believe this is actively being worked on, so please feel free to take it on.

The easiest way to set up your development environment is to use devcontainers.

Once you've spun up a devcontainer, installing the Python project for development can be done using:

cd python/cuda_cccl
pip install -e ".[test-cu12]" -v  # or test-cu13 (depending on what the container's CUDA version is)

Please, feel free to ask here if you have any questions!