The official backup tool for GitLab takes many hours to run because it
puts everything inside tarballs and then gzips each one. It seems safe
and much more efficient to skip this step for the offsite backup while
reusing the tarballs generated by the first backup to the Storage Box.

Should save ~5 hours from the borg-backup-offsite.service execution.

This is meant to address the daily HostHighCpuLoad alert triggered on
lists.archlinux.org, which due to the large number of files it has to
process (around 1.5 million). Machines with more than one virtual CPU
don't need this as Borg is currently single-threaded and thus limited
to one core.

The reason is that we might want to independently retry the offsite backup in case it fails but not rerun the primary backup in that case.
This isn't possible with the current setup and so we'll have to split this up into two oneshot services.
The way it works is that borg-backup.service will schedule borg-backup-offsite.service to run after it which ensures that
no two backups are running at the same time and also that they can be indendently retried if one of them fails.