[FEA]: Investigate remaining performance issues with PSTL algorithms

### 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.

Some of the PSTL algorithms show minor performance regressions compared to thrust algorithms.

### Describe the solution you'd like

We should investigate where those performance differences come from and see whether we can avoid them.

Some are certainly due to the usage of 64 bit indices which thrust avoids

### Describe alternatives you've considered

_No response_

### Additional context

_No response_

The issue names libcu++, PSTL algorithms, and Thrust but does not identify files, benchmarks, or specific algorithms. Start by locating the PSTL and corresponding Thrust performance tests, then compare their results and investigate the stated 64-bit index difference. Done means the remaining regressions are characterized and actionable improvements are identified.

Written by the indexing model from the issue text.