Make pipeline faster

[jlebon]

It takes about 1h right now for a full build. In the past few days, we've had to run and iterate the pipeline multiple times, and waiting for that 1h has been really painful at times. I think there are multiple ways we can speed this up:

1. first and foremost, while not related to a single build, we need to unblock concurrent builds (Allow concurrent builds #68)
2. by far the longest step is xz compression. Fortunately, xz supports -T for multi-threading. But doing this correctly for a k8s environment is tricky. See e.g. virt-install: do not assign vpu's when executed in Kubernetes coreos-assembler#284 (comment) and virt-install: do not assign vpu's when executed in Kubernetes coreos-assembler#284 (comment).
3. parallelize buildextend commands -- this will likely require driving some changes into cosa. But at least for the ones that just do gf-platformid it shouldn't be too hard. Of course, this goes right into YAML/JSON interface for COSA coreos-assembler#531; cosa itself could potentially include the parallelization code.

[jlebon]

by far the longest step is xz compression. Fortunately, xz supports -T for multi-threading.

coreos/coreos-assembler#632

[jlebon]

1 and 2 and 3 are done now. And more parallelization work is happening.