[llvm] Add constant-folding for unary NVVM intrinsics (PR #141233)

[Lewis Crawford via llvm-commits]

LewisCrawford wrote:

I would like to land this patch: https://github.com/llvm/llvm-project/pull/140270 before merging this. That patch adds an option to disable constant-folding for FP function calls, and may be useful in the few cases where the device-side execution of these intrinsics must be a bitwise-exact match to the host-side constant-folded version of the results (or when debugging).

https://github.com/llvm/llvm-project/pull/141233