Rework json parsing implementation

This merge reworks the way that the json parsing is done to provide a more uniform interface and move more of the work into the base class, which the individual parsers inherit from.

The parsing now uses a dataframe based framework. And in general the specific implementation for a given parser is both easier to implement, and the intended groupings of uncertainty_set and combine_name uncertainties is more clear.

Additionally, this naturally includes some uncertainty-specific metadata which is inferred by the name of the correction (e.g. the name of the tagger, or if it is a fixed-working point or full shape based correction)