abstract the assign and drop logic; cleanups everywhere; http endpoints for getting and assigning rules

RemoteTaskRunner#run should fail on task with taskId from tasks.keys set

1) add a log line on plumber creation to make it easier to know what rejection policy is being used.

1) Adjust RealtimePlumberSchool to take a configurable rejection policy that lets you switch between server time and message time for rejections.

1) Add new SearchQueryDimFilter/SearchQueryFilter that will do a filter using the same queries that can be used for SearchQuery

1) Fix bug with Sinks that causes Realtime it to overwrite the first chunk of data when creating a segment to hand off to compute nodes.
2) Fix bug with RealtimePlumberSchool that can cause it to miss some previously-persisted segments when directory naming changes
3) Have the various runnables in RealtimePlumberSchool name the threads to make it a bit easier to debug

2) Add GroupByTimeseriesQueryRunnerTest which wraps a GroupByQueryRunner to look like a TimeseriesQueryRunner and leverages the tests there to help verify behavior (#8)

1) Have IndexGeneratorJob write the descriptors for each of the segments it creates to a path in the temporary working directory (generally HDFS)
2) Have the DbUpdaterJob read descriptors from the temporary working directory instead of looking in the final segment output location (often the eventually consistent S3)
3) 1 and 2 Fixes #30

Fixes to indexing service to handle large merges better