Ensure long running jobs are destroyed after `kill`

[quasiben]

YARN sometimes does not properly clean up container and can leave long running processes linger. I've found a few open issues related to this:

• https://issues.apache.org/jira/browse/YARN-1842
• https://issues.apache.org/jira/browse/SAMZA-498
• https://issues.apache.org/jira/browse/YARN-318

Spark has also had issues around this in the past

• [SPARK-4117] [YARN] Spark on Yarn handle AM being told command from RM apache/spark#10129

I've verified on Ubuntu 12.04 that SAMZA-498 resolves the issue