Add Restart=on-failure for maintained systemd units to make them more robust

[invidian]

Right now, if for example Packet metadata service is not reachable, coreos-metadata will never converge, so if node reboots during this time, it will never come back.

Even when the metadata service comes back, node will still be stuck until either node is rebooted or service is manually restarted.

We should be able to automate that to make it more robust by adding Restart=on-failure.

Following https://unix.stackexchange.com/a/272650, we should probably add Restart=on-failure and RestartSec=5s (or some other value) to all units with Type=oneshot to make sure they eventually converge and not just give up.

@pothos also suggesting adding RemainAfterExit=yes:

RemainAfterExit=yes is also missing which means that currently the service is executed multiple times, each time when pulled in as wanted/required. It could mean that an existing file gets lost when the server is unavailable later (but that also depends on how afterburn writes this file out, so it may not be a problem right now).