[llvm] AMDGPU: Replace sqrt OpenCL libcalls with llvm.sqrt (PR #74197)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> So I suggest adding the first test (with fast) and removing nsz in the second. The rest seems OK.

nsz doesn't do anything for sqrt lowering. Signed zero works correctly in the instruction. We have an excess of messy and overlapping sqrt tests as is, I don't think we need to add any new ones. Adding tests with "fast" will just add to the mess. We should drop the nsz and possibly any using the unsafe-fp-math attributes 

https://github.com/llvm/llvm-project/pull/74197