[FEA]: Allow sub-partition of a block reduction

### 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.

Block reductions return a single value in the first thread, but there are instances were N results are needed in different threads. At a warp level, cub allows you to create "logical" warps, which partition a hardware thread group further, but this functionality is missing at a block level. All non-exited threads have to participate in block API call.

### Describe the solution you'd like

Two possible solutions:
- Provide segmented / batched block reduction
- Allow sub-partitions of a block

### Describe alternatives you've considered

The current work-around is to utilize cub's warp reduction and implement something custom to then combine the warp level reductions for the block but it would be nice for this to be a single cub API call.

### Additional context

_No response_