map-only job should ignore node lost event when task is already succeeded

When node lost, all mappers on this node will be rescheduled, but it's unnecessary for succeeded map only task because map only job has no reducer, and map output has been written to HDFS when it become succeeded, so we can ignore this node lost event.