[Intel-gfx] UXA clipping reduction

[Eric Anholt]

There's this odd pattern in some UXA code that appears to have been
carried over from fbfillrect.c where for each thing to be filled we:

1) clip against the clip extents
2) trivial reject
3) clip that against each box of the clip
4) reject per box
5) draw.

I don't see a justification for this complexity, in terms of how I
expect cliprects to be (generally 1 rectangle) or in terms of how I
expect rendering to be (generally within cliprects if there are any).
And, we see from the history of uxa-accel.c, we've screwed up this
clipping up before.  So, I propose instead, we:

1) Clip against each box of the clip
2) Reject per box
3) draw.

Here are the patches.  No performance testing included, but one known
bug is deleted (which I actually hope we land the small submitted fix
for first, I'm just hoping to get a name/commit message on it).