Long-Term Velox Spark Application

[tomerk]

If we can set up some sort of long-term relation between a velox cluster and a spark cluster, there are optimizations we can make such as:

• Don't keep reshipping the Velox JAR
• Only transfer over the new observations on a bulk retrain

[tomerk]

As of pull request #51 being merged, this exists. There is now room for some extra optimizations in terms of only shipping new observations.