Also use the nice VARIABLE instead of hardcoding it.

7 retries takes around 2 minutes with curl exponential delays.

Without the full name, systemctl start assumed it to be a service name.

The previous patch was wrong for two reasons:
- you can't put shell commands in a Exec, as there's no shell involved
- while you could split the two on two different Exec, it would not
  work anyhow because the unit runs as the esamesync user, and it cannot
  create the /opt/esame directory if the admin did not create it beforehand.

This one was tested to work; the machine boots correctly.

Not tested with labmanager; it is not supposed to be working properly
until you have working copy in /opt/esame, but it would be interesting
to see how it fails.

Otherwise the service will loop and a freshly installed machine
(or a machine with an empty /opt) will not finish the boot.

We should have none, but, in case there are errors reported by the WoL command,
we do not want to stop the systemd service from starting.

Hopefully this will fix the WOL issues, with machines that are sometimes
ignoring the WOL packet.

This will help troubleshooting, as the future sysadmins will connect to
the machine, run `journalctl -f ` to see what's wrong with the machine,
and in about 30 seconds they will see:

  Error fetching machine configuration!
  Error fetching machine info!
  If this is a new machine, did you add it into labmanager-admin?

So they will know what's wrong, and how to correct it.

These are not debug messages, they should be always on to see what's happening
on the machine.

Solo il software è su paperboy2 / birthdaygirl.

Nota: questo file di configurazione viene scaricato solo quando sulla
macchina non è presente. Se il file esiste, lo script non lo sovrascrive,
per non perdere le modifiche locali fatte dall'amministratore di sistema.