JobRunner.kill doesn't terminate cleanly with YarnJob.

I discovered this while working on the wikipedia tutorial with the Fluent API.

Running run-app.sh with --operation=kill will kill the job but doesn't appear to do so cleanly. There are a few reasons.
1. The client can't get the status for finished jobs because it only looks at the active (RUNNING) application IDs
2. JobRunner.kill() waits for a successful finish. It should wait for any finish
3. A killed job is considered an unsuccessful finish. It should be considered successful.