Add on to the work done in
HBASE-13992 to add functionality to do a bulk load from a given RDD.
This will do the following:
1. figure out the number of regions and sort and partition the data correctly to be written out to HFiles
2. Also unlike the MR bulkload I would like that the columns to be sorted in the shuffle stage and not in the memory of the reducer. This will allow this design to support super wide records with out going out of memory.