and point again to the upstream repo

- Do nagios near the end
- Use virsh shutdown instead of destroy.

Previous incantation of this didn't work because /root/sh got removed
and screen failed to spawn a new shell. By copying busybox in a tmpfs,
the binary is stored in core memory and therefore survives the disk
wipes.

We might not have a shell again after the long final badblocks call:
we might have lost the original terminal, network might have cut us
off, etc. By doing this in a single pipeline, we make sure the
shutdown actually gets called.