Fix the NodeLost pod's status

[abuzhynsky]

I'm aware of PR #80 but maybe it's better to filter out the pods in NodeLost status in the beginning of the workloop. And also fixing the case when the node is lost (and apiserver is potentially unreachable) but mongo instance is alive. In this case mongo-sidecar will be potentially unable to react to the events and won't add any new pods. I propose to delegate primaryWork to any other instance in this case.