Improve resilience in some failure cases (1.x)

[ghostdogpr]

• When unregistering, ping the shard manager first and interrupt the unregistration immediately if it's down (don't even stop entities in that case, because they won't be rebalanced right away).
• When a node is unresponsive, try to get the latest shard assignments from the shard manager just in case it's the storage layer that has a problem (it happened to us once that Redis had an outage and shard assignment updates were not pushed to nodes).