[X86] Don't fold AND(SRL(X,Y),1) -> SETCC(BT(X,Y)) on BMI2 targets

With BMI2 we have SHRX which is a lot quicker than regular x86 shifts.

Fixes #55138