Arvados

Add a broken-node metric

(counter) VMs that are determined to be "broken nodes"

Add a label to separate VMs marked as broken before the first container is started on them (likely boot problem) and after (likely container related problem).

Note that we already have a boot outcome metric. Make sure that we increment the broken node counter ("before first container" label) when we have a boot outcome == failed, though not in the timeout case.