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

[via llvm-commits]

Lai-YT wrote:

> Why is this a module flag, instead of a function attribute?

The other value handled by the _NVVMReflect_ pass is `"__CUDA_FTZ"`, which Clang currently sets via a module flag. I followed the same approach here for consistency.

Also, since the decision to use precise square root functions seems to apply at the module level, using a module flag feels reasonable.
@efriedma-quic, do you see any pros or cons to using function attributes instead in this case? :raising_hand: 

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