The node should shutdown if SyncingEngine event loop terminates

[dmitry-markin]

As #3496 demonstrates, when fatal error happens in SyncingEngine and its event loop terminates, the node continues running. This is not a correct behavior, as it doesn't make sense to run the node without syncing. Instead, the entire node should shutdown.

[altonen]

Maybe SyncingEngine could be started with spawn_essential() or Notifications could return a fatal error if one of the protocol command streams is closed, shutting down NetworkWorker which would probably shut down the rest of the node as well.