This is a prototype implementation, integrating StreamingKMeans clustering into
Storm as a bolt,
- StreamingKMeansBolt
and an aggregator,
- StreamingKMeansAggregator

There are 2 simple tests that create a linear topology from a spout generating
random vectors (HypercubeSpout) to the clusterer and also through some bolts
that write data to disk as sequence files.

The output centroids are then read back in and the distribution of distances is
computed and the results are reasonable (nearly equally sized clusters).