Refactor run_hackett to generalize its handling of incomplete/missing data

[djinnome]

Tensors work by indexes. Dataframes work by named rows and columns. This mismatch is never more clear than when you want to map data to a model and some of the model reactions and metabolites are unmeasured. With dataframes, it is trivial. With tensors, you need fluency in Einstein notation.

Fortunately, Peter St John has provided an example of how to do this tensor manipulation in https://github.com/pnnl-predictive-phenomics/emll/blob/master/notebooks/run_hackett_inference.py and we just need to generalize his approach beyond the specific example it was used in.

[augeorge]

related: pnnl-predictive-phenomics/syn_bmca#3