[clang] [llvm] [Clang][NVVM] Support `-f[no-]cuda-prec-sqrt` and propagate precision flag to `NVVMReflect` (PR #134244)

[via llvm-commits]

Lai-YT wrote:

Got it. I did come across `nvptx-prec-sqrtf32`, but since _NVVMReflect_ makes the decision earlier by selecting the intrinsic in _libdevice_, I figured I couldn't rely on it.
I completely agree that unifying the behavior would be ideal moving forward.

Thanks so much for the insights, @Artem-B and @AlexMaclean! 🤗
It sounds like there's not much more I can do with this PR as it stands.

https://github.com/llvm/llvm-project/pull/134244