This way the individual instruction handlers don't have to bother about the
color correction, and there's now a single place to implement fixup dependent
sampling writemask selection.

The color correction cannot be done behind the back of the individual
instruction handlers because it might conflict with the instruction's
color modifications and the D3D provided writemask.

Based on a patch by Aric Stewart.

This is done by wrapping them inside an empty 32-bit module.