[FEA]: cub::DeviceSegmentedSort with decomposer

### 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

cub

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

I would like to be able to run `cub::DeviceSegmentedSort` (and/or `cub::DeviceSegmentedRadixSort`) with a non-primitive type, ideally through a decomposer as in [`cub::DeviceRadixSort`](https://nvidia.github.io/cccl/cub/api/structcub_1_1DeviceRadixSort.html#id1).

### Describe the solution you'd like

Decomposer support added to `cub::DeviceSegmentedSort` and/or `cub::DeviceSegmentedRadixSort`. Thank you!

### Describe alternatives you've considered

We've tried running DeviceRadixSort with the decomposer on our list of `(i, k)` tuples (the list is already sorted by `i` though), but this is a significant bottleneck.

### Additional context

_No response_