[FEA]: Add verification to TopK tuning

### 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.

TopK tuning is missing verification of the algorithm output:

https://github.com/NVIDIA/cccl/blob/5ea5d42567693aad8c0e5d6316155ecc612fdc71/cub/benchmarks/bench/segmented_topk/variable/keys.cu#L60-L77

### Describe the solution you'd like

Before integrating verification into all the tunings, we should prototype its shape on one algorithm without making significant changes on the infrastructure side. Let's iterate on segmented TopK verification first.

This issue can be closed by a PR that:
- adds verification that terminates benchmark in case the algorithm produced incorrect output
- modified benchmark [overview](https://github.com/NVIDIA/cccl/blob/main/docs/cub/benchmarking.rst) with new practice

### Describe alternatives you've considered

_No response_

### Additional context

_No response_