lab factory should collect crash dumps from machines that panic

[jclulow]

If a machine running a lab job panics today, the console output during the panic (which usually includes a stack trace) is included in the event stream for the job by the lab factory. In the environment provided by the agent, though, we do not configure a dump device -- and even if we did, there is no subsequent factory-controlled boot to get in there and recover the dump.

We should add support for configuring a dump device, and for booting from a maintenance image (different to the target-specified image) for dump recovery.