[llvm] [AMDGPU] Gate rootn(x, +-2) -> sqrt/rsqrt fold on nsz/ninf (PR #200578)

[Matt Arsenault via llvm-commits]

https://github.com/arsenm approved this pull request.

Could do better with computeKnownFPClass but that's extra testing work 

https://github.com/llvm/llvm-project/pull/200578