[Intel-gfx] 965 check_aperture/batch_start_atomic

[Eric Anholt]

Here's what I've been working on for converting our 965 3D code to
correctly using check_aperture and batch_start_atomic.  This should fix
stinking piles of races/hangs/misrendering that would have been hard to track
down before (as we experienced last week with 915-class).

The series appears to work up until the last render commit, which I'm now too
bleary-eyed to spot the fail in.

Also, writing 3 965 3D rendering pipeline implementations sucks.  Can I get
a credible X-on-GL already?