[llvm] [IVDescriptors] Don't require nsz/nnan for (min|max)num. (PR #137003)

[Andy Kaylor via llvm-commits]

andykaylor wrote:

> 2\. we can avoid the problem for the purposes of lowering C's fmax/fmin without no-honor-nans by multiplying by 1.0 before doing fmaxnm (the multiplication sends sNaN to qNaN, which will then be dropped when the other operand is a number).

Multiply by 1.0 is likely to be optimized away. I think you'd need `llvm.canonicalize` instead.

https://github.com/llvm/llvm-project/pull/137003