[llvm] [X86][DAG] remove LowerFCanonicalize (PR #188127)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> Second, the absence of the optimization `x * 1.0 -> x` for STRICT_FMUL is not an intended behavior, it is a deficiency of the current Strict FP mode implementation.

Yes it is intended behavior. There's a narrower range of situations where it's legal in strictfp to eliminate this. You would need to prove that the value isn't a signaling nan, and isn't a non-canonical value 

https://github.com/llvm/llvm-project/pull/188127