feat: run jobs on every tf apply

[glasnt]

Requires GoogleCloudPlatform/avocano#164 to be merged, released, and the new version referenced

Resolves #44

Cloud Run jobs are run through the metadata startup script on the GCE instance. Previously, when this terraform was first applied, this gce instance would stay on, and subsequent terraform apply commands would only alter the state of the job or gce instance, but would not restart the instance, so the job execution would not be performed again.

With this change, after the gce instance has completed it's tasks, it shuts itself down. The TF designed state is running, so any time terraform is run, the instance will be restarted, and the jobs re-run. The change in avocano now ensures all these jobs are idempotent (previously, re-running the setup job would create duplicate products).

This change now reasserts server image, client deploy, and database migrations.

(Note: the setup script includes base migrate/collectstatic commands, as well as database seeding. It's that seeding that's now idempotent)

[grayside]

I raised a couple non-blocking out-of scope points, and some questions that are more about my learning the project than any challenge to this change. I've prefixed review comments based on https://conventionalcomments.org/ to help make the context of each point explicit.

LGTM and merge when ready.