Refine job status: Unknown pod

[hao1939]

Features: if node loses, mark job as "Unknown" or "NotFound" (dependent on the pod status), then resubmit.

Known issues:

1. May need to reduce the timeout. It takes about 7 minutes before pod turns into "Unknown", we may don't need to wait so long. (Also notice, if we reduce the timeout, we would be more likely to kill the job by mistake if there are problem on kubelet.)
2. If node resume before we resubmit the job (it takes about 10 minutes or more), the job will end up with status 'failed'.