Make repl send_event idempotent and retry on timeouts

If we treated timeouts as failures on the worker we would attempt to
clean up e.g. push actions while the master might still process the
event.