[Mesa-stable] Please revert "nv50/ir: normalize cube coordinates after derivatives have been computed" in stable

[Ilia Mirkin]

Hi Emil,

Looks like commit 6eeb284e4f74a2fe5ae6cba90f97f219935e24df causes some
regressions in the tex-miplevel-selection test, even though it fixes
some dEQP tests. I don't know why this is happening... I'm kinda
suspecting that the tex-miplevel-selection test is wrong, but it *is*
passing for all the other drivers, so I think we should just revert it
in stable for now while I sort out what's going on.

[Looking at some other regressions right now... will let you know if
there are more reverts/cherry-picks required.]

Thanks,

  -ilia