[llvm] Revert "Revert "LangRef: Clarify llvm.minnum and llvm.maxnum about sNaN and signed zero (#112852)"" (PR #170067)

[Ralf Jung via llvm-commits]

RalfJung wrote:

This revert re-introduces the bug that the x86 backend is wrong. @arsenm what do you propose a frontend should do if it wants IEEE754-2008 `minNum` (*not* IEEE754-2019 `minimumNumber`) semantics? For many years, that's what `llvm.minnum` did, and so that's what frontends like Rust want. IOW, this docs change is a breaking change. The 2008 semantics are also what many LLVM backends and optimizations do when they see `llvm.minnum`. It's not enough to just change the docs, you need to change optimizations and backends as well -- and from all the discussion that has been had, it's not clear that there is consensus to do that. Furthermore, people have waited many months since the first doc change, and there has apparently been zero progress towards actually implementing the new semantics in backends and optimizations.

https://github.com/llvm/llvm-project/pull/170067