[FEA]: Investigate performance regression with `cuda::std::partition`

### 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

libcu++

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

We see small performance regressions for large arrays in #8081 .
```
| I8 | 2^24 | 0 | 90.017 us | 1.32% | 97.427 us | 2.29% | 7.410 us | 8.23% | SLOW |
| I8 | 2^28 | 0 | 1.270 ms | 1.13% | 1.361 ms | 1.52% | 91.787 us | 7.23% | SLOW |
```

### Describe the solution you'd like

We should see whether we can avoid those regressions

### Describe alternatives you've considered

_No response_

### Additional context

_No response_