Rationalize Calibration Calculation and I/O

[joezuntz]

Currently the various CalibrationCalculator classes in utils/calibration_tools.py which compute the shear calibration are all created and run by the source selection subclasses in source_selector.py, which gradually load and add data to them. When this is complete and the calibration factors can be calculated, the section classes call the collect methods on the calculator classes, which returns calibration information (and number count info).

Right now, the source selector classes do the work of saving that calibration information to an HDF file group. Later, when it is loaded back in to be used, Calibrator classes from utils/calibrators.py are created from that saved file info by the BinStats object.

It would make more sense to have the Calculator classes return BinStats instances directly, and add to the Calibrator classes methods to calibrate variances instead of shears themselves. Then we could combine the many compute_output_stats methods in the selector classes, and also greatly simplify MeanShearInBins.collect.

We could also think of a better name than BinStats