This is the default hash-ish shuffle (supported by SparkSQL) using serialized objects when there is no need for map-side aggregation. Part of #9
