HTMRunner only runs when generating network data

[BoltzmannBrain]

With the param generateData set the HTMRunner will create self.dataDict as it reads in the CSV and writes CSVs formatted for the network API. If you pass an already formatted data file to the HTMRunner (i.e. as dataPath) it will skip over the creation of self.dataDict. This causes experiments to fail in HTMRunner.mapLabelRefs().

@subutai I'm holding off on addressing this issue because (i) it's not blocking anyone (just causing the NLP integration tests to fail), and (ii) changing the data input format to the more agnostic version we discussed (i.e. models work with a data structure, not CSVs) then this issue would become obsolete anyway.