[llvm] [LangRef] Clarify specification for float min/max operations (PR #172012)

[Ralf Jung via llvm-commits]

RalfJung wrote:

We can do either. Is there one option that's better for backends / follow-up optimizations? What other operations do (`pow`) is to fold all NaN the same way -- that also has the advantage that one can fold without knowing which kind of NaN it is.

We also need to ensure that scalar evolution is aware of all this non-determinism and doesn't try to predict what these operations do.

https://github.com/llvm/llvm-project/pull/172012