[llvm] [DAGCombiner] Eliminate fp casts if we have the right fast math flags (PR #131345)

[Harald van Dijk via llvm-commits]

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

[Outdated documentation](https://llvm.org/docs/LangRef.html#fast-math-flags) means that this optimization intentionally contradicts documentation which says `fast` implies only `nnan`, `ninf`, `nsz` because those are the only fast math flags. That shouldn't prevent this going in, but it would be good to get that documentation updated too.

https://github.com/llvm/llvm-project/pull/131345