Also fix some minor typos etc. along the way.

Previously the --vm option, which got replaced in commit bf88ce21, would
change the install disk from /dev/sda to /dev/vda (intended for our
previous libvirt setup).  This more generalized form has no current use
case in our production environment, but may be used e.g. for testing
with local VMs or other situations.

This reverts commit 7b7035ca.

We don't have any Linux hosts (VMs) with a serial console anymore. Also
enabling serial-getty@{ttyS0,ttyS1}.service on hosts without a serial
console spams the system log with error messages. (This could probably
be fixed with small adjustments to the unit, but it's superflous now
anyway.)

Otherwise, ansible-pull doesn't work because karo is offline right now

Between ansible-pull (including pacman -Syu) and the reboot, the kernel
may fail to load modules (if Arch package 'linux' was updated), services
may run with old configuration etc. We do a reboot every night anyway,
so just move ansible-pull closer to it.

See also #6

https://jinja.palletsprojects.com/en/2.11.x/templates/#whitespace-control
Adding a "+" did not work, so I moved the statement blocks to separate lines.

The VMs will be hosted at the IRB now.

Running domains can be automatically suspended/shutdown at host shutdown using the libvirt-guests.service systemd service. This same service will resume/startup the suspended/shutdown domain automatically at host startup.

Otherwise the inventory script will stacktrace while parsing !vault tags

Explain the non-default behaviour in ansible.cfg and add references to
the deprecation discussion.

Fixes #8

Adds ordering to ansible-pull.service and periodical-reboot.service to prevent overlapping.