Optimise CI/CD run times

Our CI pipelines take quite a while, spending a long time in the setup process - any optimisations we can do would be quite helpful.

Listed in order of lowest to highest hanging fruit:

• Only run the container tests on CI (Vagrants are a bit flakey not to mention slow)
• Possibly run containers and k8s in parallel (if we have enough GHA capacity)
• Build the CMOS image once and reuse it (perhaps as a job that runs before all others and saves it as an artifact or pushes to GHCR)
• Run only smoke tests on PRs, and run the full integration suite overnight on a cron timer