JIT: Call gtFoldExprConditional in if-conversion and allow float SELECTs

[BoyBaykiller]

• Allow SELECTs for floats (only temporally for optimizations, not into codegen)
• Call gtFoldExprConditional (this handles simple folding for int + float)
• Add missing GT_SELECT equality (without this gtFoldExprConditional asserts)
• Move the 32-bit bailout to after SELECT optimizations so that it can still profit from them. Only when there were no optimizations done abort

[dotnet-policy-service]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

[tannergooding]

Allow SELECTs for floats (only temporally for optimizations, not into codegen)

There's notably no reason to not allow it into codegen if the input is a floating-point comparison.

Such cases can emit cmpss followed by blendvps, which is smaller than ucomiss; test; jcc; etc

[BoyBaykiller]

That'd be nice to handle. Currently doing that hits assert isGeneralRegister(dstReg) && isGeneralRegister(srcReg) in emitxarch.cpp:7955 so that probably needs some extra work (different PR)