So I can see which host I'm deploying to right now.

Using the forward script. It turns out to be much more reliable (and
faster).

Only externally (the runner still uses the same environment variable). I
now have it in the .bashrc.private file so it needs to be more specific.

Using Docker Compose over SSH causes errors, but it's handled. The
tracebacks just make it harder to follow what's happening.

Seperate the runner and jobs to a network just for themselves.

Select the host in the deploy script, default to all hosts.

The issue I'm trying to address here is that because of the Docker
volume, /builds/foo is really /var/lib/docker/volumes/123/_data/foo.
This causes some interesting issues with addressing files with Docker
from inside a Docker container. Do it just on ns4 (for now, if it works
and needed I may do it on host01 too). Also, deploy to the kodi host
last because it has issues right now.

- Use CI templates.
- Use common pre-commit config from snippet.
- Address pre-commit issues.

Can't deploy with the runner, the jobs fail because the runner is
restarted during. Do the pull and build with the CI, but the actual
deployment with a script locally.

Can't use the "host" variable as the tag. Workaround that with YAML
anchors (ugly but works).

- New version.
- pre-commit.
- GitLab CI/CD.
- Always pull new images.

Needed for easier mounting using `--volumes-from` to acces the files
when running another container (launching a container from inside a
container, for an example see the CI from
https://git.shore.co.il/nimrod/resume).

Set the runners to run untagged jobs (I only enabled tagging to run
specific jobs in specific runners, the untagged jobs should still run
wherever).