[SPARK-3072] YARN - Exit when reach max number failed executors

In some cases on hadoop 2.x the spark application master doesn't properly exit and hangs around for 10 minutes after its really done.  We should make sure it exits properly and stops the driver.

Author: Thomas Graves <tgraves@apache.org>

Closes #2022 from tgravescs/SPARK-3072 and squashes the following commits:

665701d [Thomas Graves] Exit when reach max number failed executors