ci: Use new 64-core runners for swrast testing.
About the week before last, we ran into some trouble getting code merged because fd.o's shared runners were backed up, so I've arranged for Google to host 2 64-core ICL machines to take over the swrast testing.
Before, we had 26 swrast jobs being spawned in a full pipeline, not counting the many build and container jobs running on the x86 runners. fd.o has 3 shared runners on packet.net with I 8 slots each, so unless gstreamer had unused capacity on its htz runners, we wouldn't be getting our pipelines through in the time we were hoping for even if they were all <10 minutes. And that's assuming that nobody else was running anything on fd.o CI at the same time.
Now, the bulk of the swrast jobs are moved to 2 runners each fully dedicated to the job they're running while they've taken it. We've got about 30 minutes of runtime on them for a full pipeline, so about 15 minutes each. This is still not the 10 minutes I was originally aiming for from the driver testing stage, but I think now hardware testing will become the long pole again. Plus it'll free up shared runner capacity for doing our build jobs.
There's about 1 minute of overhead per CI job where the workload is not massively parallel (container setup, pulling artifacts, uploading results, etc.), so I've pulled piglit into the testsuite jobs as well. I consider this a win anyway, now for all of (say) softpipe's results, you can just go look at the job named "softpipe" (well, except for asan, because asan artifacts are in a different tarball so you can't run them both from the same testsuite)