WIP: llvmpipe: fix 4-bit output scaling + polygon offset.

This is overkill, but hey 4-bits per channel is hardly something to care about.

(Suggestions welcome for a better version).

Fixes: dEQP-GLES2.functional.fbo.render.rgba4

@sroland this is more of an RFC in patch form.

I've added a second RFC patch for polygon offset fixes.