[Piglit] Render-to-FBO and concurrent shader_runner.

[Eric Anholt]

With all the new tests we're adding, I was feeling guilty about our
piglit runtime.  I figured that we could run tests into FBOs and avoid
a bunch of overhead in talking to the X Server and window manager
about window setup.  That doesn't appear to have been a limiting
factor (didn't do stats, but didn't look like a statistically
significant difference after converting), but I think it's worth it
anyway for not popping up stupid windows as much.  And, now that we're
rendering offscreen, I feel like running things in parallel even if we
use the GPU is worth it -- our test runtime is CPU limited, not GPU,
and anything we can do to make testing easier should help driver
quality.

This patch series reduced runtime from 11 minutes 47 seconds in the
before run to 6:38 in the after run on my dual-core system.

For those of us working on temporarily unstable drivers that impact
other tests when run in parallel, we should be able to just use -c
off.  And, as Ken noted, it's not like the X Server isn't running the
GPU in parallel with our GL drawing already anyway.