[FEA]: Provide CPU backend for cuda.compute algorithms

### Is this a duplicate?

- [x] I confirmed there appear to be no [duplicate issues](https://github.com/NVIDIA/cccl/issues) for this request and that I agree to the [Code of Conduct](CODE_OF_CONDUCT.md)

### Area

cuda.compute (Python)

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

Dear CCCL team,

In the scope of the Awkward Array project it would be great if it is possible to dispatch to CPU kernels as well. That would allow us to maintain a single set of kernels (independent of the accelerator) and decide at runtime to which accelerator we want to dispatch to.

Is this possible, or are there plans to make this possible?
(Also: if this is possible, does this also support kernel fusion?)

Best, Peter

cc @ianna

### Describe the solution you'd like

As described above we'd love to be able to decide upon 'compute' which accelerator to run on.

### Describe alternatives you've considered

_No response_

### Additional context

_No response_