Transient failures in instance state monitor cause instances to be lost at sea

Sled agent spawns a state monitor task per instance that calls Propolis's instance_state_monitor endpoint, watches for VM state transitions, processes them, and relays any resulting instance state changes to Nexus. Any error in Instance::monitor_state_task or any of its callees bubbles up through the task and causes it to exit. After this, nothing monitors Propolis for subsequent state changes or notifies Nexus if they occur.

The DNS resolution error in #2726 brought this to my immediate attention, but any failure in the monitoring task will do the trick; another not-too-esoteric one is a Propolis panic that causes the call to instance_state_monitor to fail.

Start in the Sled agent at Instance::monitor_state_task and trace its callees to the instance_state_monitor endpoint, including how errors exit the task and how changes are relayed to Nexus. Reproduce or test a transient monitor failure, then verify that later Propolis state transitions are still observed and reported to Nexus.