[Intel-gfx] intel-cleanup branch

[Eric Anholt]

Cutting 10% of the total lines of code in the 2D driver seems like a good
step towards reducing our maintenance burden.  I still find new things to
remove every time I look.  But more importantly, the number of different
combinations of paths goes way down -- XAA accel was way different from EXA,
was way different from UXA, and XF86DRI has a whole bunch of ways to fail
that's gone in DRI2.

That does mean that we need to be sure we get the regressions in the current
stuff fixed.  In particular, vblank syncing with DRI2 (I think we've got
a coherent plan here now), and supporting the fail-at-initializing-DRI path
for the 2.6.28 kernel with a second X Server.  I'm working on this second
one, and I think jbarnes is doing the first.

Long term, we're hoping to start relying on KMS around 2.9 (fall) or 2.10
(winter) after we've made sure that we've cleaned up known regressions there.
That'll cut some 15,000 lines of code, or about a third of what remains after
this series.