The fixed function backend disables the extension. Also messing
with the final combiner shouldn't be needed if nvrc is disabled.

This is a nop by itself, but it makes it easier for fragment pipelines
to read WINED3DTSS_TEXTURETRANSFORMFLAGS.

There's no point in separating scale and luminance since they're
uploaded at the same time. We still want to separate the stages
though.

Note that the fix is only implemented in the arbfp and atifs fragment
pipeline. Unfortunately nvrc doesn't support dot3 as alpha operation.