changes of how series are handled, including implementing GWFSeriesReporter

mucked around with how series are generated in classifier.timeseries to support a slightly different data structure. Modified calibration.series2dict to reflect this along with io.HDF5SeriesReporter. Implemented GWFSeriesReporter. I also identified an issue with the beta distribution approximation for the calibration pdfs, which was causing problems when the pdfs are vanishingly small. There's a bit of a hackey fix in there now, but it should serve.

I tested this by running the entire batch pipeline and then confirming that I am able to read in series data from disk using either HDF5SeriesReporter or GWFSeriesReporter.