Refactor hazard calculation pipeline to keep ERF logic trees together

Background: nshmp-haz-v1 was structured to run so-called SourceSets; logical groups of ruptures that were 1:1 with XML (formerly fortran) input files that had a lot of redundant information in them. In refactoring to JSON/CSV formats, models/ERFs were organized into SourceTrees. At the tip of a SourceTree branch there is a RuptureSet. RuptureSets replaced SourceSets in calculation pipelines, and while RuptureSets carry enough information to be able to recombine results into mean hazard and disaggregation results (and hazard and disagg results decomposed by SourceType and GMM), the original SourceTree info is lost. In order to support improved organization of disaggregation contributors and output component curves for uncertainty analysis, we need to preserve SourceTree groupings through various calculation pipelines.