Investigate CUDA 12.8 `nvcc`/`nvrtc` compile time tracing

[ptheywood]

Some of our compilation unit's are rather slow / time consuming to compile.

With CUDA 12.8, compile time tracing was added to nvcc and nvrtc.

It may be worth at least a quick check to see if there are any easy ways to save time here. (nvrtc improvments would be more impactful, but for 12.8 most of the nvrtc time is jitify so still dependent on #1150)

https://developer.nvidia.com/blog/optimizing-compile-times-for-cuda-c/

[Robadob]

Jitify limitation is just redundant repeated header parsing, so I don't think we'll learn anything on that front, already done best we can to address it (providing all known headers up front).