[llvm] ValueTracking: llvm.amdgcn.fract cannot introduce overflow (PR #189002)

[Yingwei Zheng via llvm-commits]

================
@@ -63,4 +63,30 @@ define double @ret_fract_no_nan_no_inf(double nofpclass(inf nan) %x) {
   ret double %fract
 }
 
+; Fract must be <= 1, so this cannot introduce overfloww
----------------
dtcxzyw wrote:

```suggestion
; Fract must be < 1, so this cannot introduce overflow
```
See https://github.com/llvm/llvm-project/blob/e55c85341d9f757021b4f31b910d2ad2175f7e6c/llvm/lib/Analysis/ConstantFolding.cpp#L2553-L2564

https://github.com/llvm/llvm-project/pull/189002