Improve robustness and usability of RealtimeIndexTask and some code cleanup in the Indexing Service

Minor fixes - maven dependency, unit test that was failing and quirk for third party S3 provider

Fix issue with realtime nodes selecting intervals that are too large in the results

Adding a dependency on Apache commons httpclient in the Druid client pom.xml. It currently gets pulled in through some dependencies but it should be explicitly set.

- Realtime plumber will start persisting segments shortly after finishJob is called, regardless
  of rejection policy

- Add "none" rejection policy

Changed method of publishing segment ownership to ZK to use batching, includes changes to allow for migration