Fixes issue #75.

This is just adding complexity, instead define the variables directly
in the run script.

Also fixes an issue where containers were not restarted when the
environment changed.

prometheus: account for one-minute timers when detecting crash loops

See merge request !100

The ballpark upper limit is systemd's RestartSec. The default is 100ms,
or 6000 maximum restarts per 10 minutes.

The probe attempts an unverified SSL connection, so it should work
even before ACME gets real certificates.

Without 'state: directory' a task would fail on initial setup.

Allow running NGINX on a different (local) port

See merge request !99

The podman-pull-image command is renamed float-pull-image and it is
meant to work with Docker too: the docker_image Ansible module does
not actually generate a status in a way that is useful to us.

base: add ipmi support and autodetection

See merge request !98

In advance of moving fully to Podman, which does not have a server
component in the first place. We are not using the Docker metrics
anyway.

Otherwise we risk referencing a source repo that does not exist, as
is the case for stretch-backports during a stretch -> buster upgrade.

docker: stop leaking password on command line

Closes prod#150

See merge request !97

Fixes ai3/prod#150

Add base playbook

See merge request !95

Python 3 fix, can't compare None with 0.