Without this sanitizing it's possible to perform a shift with a shift
amount that's the same size as the type being shifted. This actually
occurs when decoding ORR variants.
We could get fancier here and make this branchless, but we don't
really use RotateRight in any performance intensive areas.