[FEA]: implement virtual shared memory in c.parallel and cuda.parallel merge_sort and unique_by_key

### 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.parallel (Python)

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

`cub::DeviceMergeSort` uses virtual shared memory in order to support larger data types. This is currently not implemented in c.parallel and cuda.parallel, see https://github.com/NVIDIA/cccl/blob/5857e3d0893e9767504a9944da3d5b7ea82b4ff5/c/parallel/src/merge_sort.cu#L261-L265

### Describe the solution you'd like

Implement virtual shared memory for large data types

### Describe alternatives you've considered

_No response_

### Additional context

_No response_