Fixed bug where added cofactor reactions would get deleted

[jonstrutz11]

When we remove reactions due to cofactor redundancy (Pickaxe._remove_cofactor_redunancy()), the following code can delete reactions created for previous cofactors in this for loop (for cpd_id in cofactors_as_cpds:).

Specficially, it can add them to this set on subsequent runs of the loop:

rxn_ids = set(
                self.compounds[cpd_id]["Product_of"]
                + self.compounds[cpd_id]["Reactant_in"]
            )
rxns_to_del = rxns_to_del.union(rxn_ids)

The code that creates new reactions is further down in _remove_cofactor_redunancy():

        # Reaction does not exist, generate new reaction
        else:
            cofactor_rxn = {
                "_id": cofactor_rxn_id,
                # give stoich and id of reactants/products
                "Reactants": [(s, r["_id"]) for s, r in reactants],
                "Products": [(s, p["_id"]) for s, p in products],
                "Operators": rxn["Operators"],
                "SMILES_rxn": rxn_text,
            }
        
            # Assign reaction to reactions dict
            self.reactions[cofactor_rxn_id] = cofactor_rxn

So even though this new reaction is created, sometimes it will then get added to the rxns_to_del set the next time the loop runs and will get subsequently deleted. I think this only happens if there are > 1 cofactor on one side of the reaction (so when the loop gets to the other cofactor it sees this reaction as one to delete), but I'm not 100% sure on this.

See commit changes for proposed fix. Basically I just make sure to not add any of the newly created cofactor reactions to the rxns_to_del set in subsequent runs of the loop.

This fix seems to increase final number of predicted reactions by around 3% and compounds by around 2% in my use case (and those 3% added reactions all make sense and I would expect them to be in the final set).