Fix NPE in RemoteTaskRunner when some tasks in ZooKeeper but not in Overlord

[jihoonson]

Fixes #5506.

---

This change is 

[clintropolis]

👍 We should call out in release notes that there will be a small per task increase in the amount of data stored in zookeeper.

[drcrallen]

If the task has already been shutdown, will this make it start back up again?

[drcrallen]

I guess there's already a "we'll add it if we don't know about it" thing immediately below, so the check must be elsewhere.

[drcrallen]

@jihoonson can you record in this PR where that assurance is made... prior shutdown tasks should be terminated instead of added to the active queue. I forget where it is and its not easy to find.

[jihoonson]

@drcrallen if you mean TaskQueue by the active queue, TaskQueue is a waiting queue for tasks. When a task becomes ready, it calls TaskRunner.run() to run a task. Once a task is added to TaskRunner, it isn't added back to TaskQueue.

When the remoteTaskRunner sees an unknown prior shutdown task, it adds the task to its runningTasks (https://github.com/druid-io/druid/blob/master/indexing-service/src/main/java/io/druid/indexing/overlord/RemoteTaskRunner.java#L966-L983) and immediately completes that task (https://github.com/druid-io/druid/blob/master/indexing-service/src/main/java/io/druid/indexing/overlord/RemoteTaskRunner.java#L994-L996).

[jihoonson]

there will be a small per task increase in the amount of data stored in zookeeper.

@clintropolis thanks for the review. This also sounds good.

[jon-wei]

@drcrallen did you have any more comments on this PR?