Reducer class reduce the (baseid, List<(vid, vdata)>) into a dictionary
(lvid, vid), and a new vertex data file (lvid, vdata), where lvid = baseid +
vid.index * splitsize; Because the splitsize is fixed for all mapper using
option mapred.line.input.format.linespermap in the JobConf,
this guarantee all vids are mapped into [0, ..., |V|-1].