investigate how to use `llvm.is.fpclass` and replace `int_dx_isinf` with it

[farzonl]

HLSL Intrinsic	DXIL Op	LLVM Intrinsic
[isinf]	[IsInf][dx.IsInf]	[llvm.is.fpclass]

Originally posted by @bogner in #87367 (comment)

Justin's lowering table shows we don't need our own intrinsic here

[damyanp]

@farzonl - is this still needed or are there other issues tracking this?

[farzonl]

@damyanp This is still need and will likely impact the design of #137209

[Icohedron]

@tex3d mentioned that the isinf and isnan ops are intended to not be removed when using fast math.
Therefore, we shouldn't use the builtin llvm.is.fpclass because it will be removed by fast math optimizations.
Having a dedicated intrinsic (such as int_dx_isinf) prevents fast math optimizations from removing it.