sled-agent failed to notice propolis-server panicked

The propolis-server running an instance on rack2 panicked, but the instance state continued to be reported in the console as "Running". The instance_state_monitor task in sled-agent is responsible for observing state changes from propolis, which loops calling the instance_state_monitor endpoint. If propolis-server panics, then propolis will stop returning requests from that endpoint. I am unsure of what happens in this case; I assume the request will timeout or fail and report an error, which will exit the task and result in a similar failure mode as #2727.

Start with sled-agent's instance_state_monitor task and trace what happens when the propolis instance_state_monitor endpoint stops responding after a propolis-server panic. Compare the behavior with the failure mode in #2727, then establish tests showing that a panicked server does not leave the instance reported as "Running".