Fix the case where a task completes during indexing service coordinator restart causes improper cleanup

1) Fix bug with NPE thrown when requesting a dimension on a GroupBy query that is not dictionary encoded

1) Fix bug with SingleSegmentLoader.StorageLocation keeping track of its storage size incorrectly.  Add unit test ftw.

Improve runtime of segment distribution algorithm

Kafka protobuf

Robustness fixes for properly keeping track of when Segments are accessible

1) Add check in ServerManagerTest to make sure that the Segment has been "checked out" before the factory ever sees it.
2) Some code readability changes to ReferenceCountingSegment

update the server manager unit tests to check for the case where a segment may be accessed in toYield calls before an increment occurs