[llvm] [SDAG] Drop select -> fmax/min folding in SelectionDAGBuilder (PR #93575)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> > No. We at least should have the nnan cases to minimum/maximum combine handled
> 
> @arsenm Ping me after most of the regressions are addressed :)

These are only theoretical regressions and I don't think should hold this up. It would only be helpful for targets that 1. only have legal fminimum/fmaximum and no legal fmin/fmax, and also those patterns appear during legalization. I believe wasm is the only relevant target for 1 

https://github.com/llvm/llvm-project/pull/93575