Consider using netboot for automated recovery

[stapelberg]

We’re writing images to the SD card instead of using netboot so that the system under test models the real-world production systems as closely as possible.

Perhaps we could still use netboot in order to recover a botched test run, though: if we served DHCP+TFTP, we could serve a kernel+initrd which would store a known-good image on the SD card.

Given that this requires considerable effort, let’s wait with the implementation for a while and see how often manual recovery is actually necessary with the current model.