[llvm] [AMDGPU] Fix fmed3 constant-fold sign-of-zero miscompile (PR #201896)

[Wooseok Lee via llvm-commits]

wooseoklee wrote:

> Can this bug be hit with fmed3(N, +0, -0) where N is non-zero?

If N is non-zero, the existing compare() can handle comparison between N and +/-0. For the remaining +0 and -0, it can be well-handled by the new lambda function, so I think it is fine. I can add more test case for that.

https://github.com/llvm/llvm-project/pull/201896