Draft: First attempt at constituent loaders for soft leptons
Modify electron loaders written by @dkobylia and add analogous muon loaders. The immediate goal is to add bJR4v00 small-R jet regression, which uses soft lepton features, but should be generic enough to accommodate future models seeking to add lepton constituents.
Additional info:
The b-jet calibration group has trained a new version of the small-R regression algorithm that handles the soft-tagged muon and electron as constituents in new collections (each nominally of size 1). See this presentation for more info. I am now trying to extend Dmitrii's MR (see !78338) to accomplish this. I should note that I have not yet attempted to compile it, since I still am not sure the best way to do several things.
For completeness, the set of soft muon variables are the SMT inputs (the output score is not used) and several electron variables defined here (I believe originally by Dmitrii as well). We used TDD in release 23.0.26 to produce our training datasets, as a reminder. I attach the full set of input variables to the model. bJR4_variables.txt