monitor actor restarts

[vigdorchik]

Too frequent restarts almost surely signal some user error, affect latency any way. There must be a way to count those in a supervisor. There should also be a way to tell what actor class crashes, in case multiple are supervised by the same supervisor.