The --mount option is more strict that the --volume option, and will
error if the host source directory does not exist, whereas --volume will
silently create a directory.  We want the strict behaviour, so this
changes to use --mount.

The motivation for this change is to support job-runner running in
docker, which requires that the paths it uses are correct from the
host's perspective.  Currently, a incorrect path will be sort of work,
as the directory would be created inside the container, but only later
would the issue arise.

https://docs.docker.com/engine/storage/bind-mounts/#differences-between--v-and---mount-behavior

Was slow, complex, and unused

This simplifies the docker container by removing all the previous
user/group management inside the container as it is not needed. Instead,
we used the much simpler solution of adding an additional group to the
running user that is the host's docker group id. This allows the
container to have permission to access the host's bind-mounted
/var/run/docker.sock docker, but the container itself can run as the
local running user, which DTRT with respect to file permissions on bind
mounts.

The job-runner container uses the host's docker service. This means that
when it bind mounts directories with `docker run`, the source path must
exist on the *hosts* file system.

We've fixed this already for tests, with PYTEST_HOST_TMP, but we had not
actually fixed it for actual using the docker image for real. Before the
previous change to use --mount rather than --volume, a job would
succeed, but the files would be in the wrong location. Now we have that
change, the job fails.

We use a simple if inelegant fix for this: we mount the STORAGE_BASE
directories inside the container at the same path as on the host. This
means that the same config value can be used for job-runner copying
files itself, or for passing the host's docker to mount.

As part of this, I changedthe way we supply these values to being in
one time generated docker/.env file. This is more discoverable, usable
outside of just, and more efficant than before. I chose to use
`docker/medium` and `docker/high` as the volume mount locations, for
simplicity, and also, so the the docker machinery is isolated from the
non-docker venv set up.

This include a basic `just docker/functional-test` command to spin up
a job-runner container and run an actual job on it.