When the flag is on, stopped instances are still discoverable.
When a CN goes down, instances are still considered "running", are still discoverable. Fetching metrics will return a 500 due to timeout on cmon-agent. Prometheus will report up=0 in this case. This behavior is unchanged.
Fixes #36