Fail startup/network timeout earlier. Exit whole test suite.

[selfhoster1312]

This greatly simplifies the VM starting logic:

• only try to start the VM once... why would it work after a failure when repeating the same steps?
• apply a hard 10s timeout (with SIGKILL) on attempts to check if systemd has booted and the network is available ; this means the initial setup step should at worst fail after ~100s (5x10 + 5x10)
• fail this entire _LXC_CREATE initial step and therefore the entire test suite if bringing up the VM failed... as the message used to say, it failed miserably, and so there is no reason to keep going and run into weirder and weirder errors such as the instance name not existing because it was not created

Works for me™

[selfhoster1312]

So it turns out those timeouts are too low in real life conditions... I don't understand why/how LXD can be so slow on a recent machine?!