`PersesAtomMapper` combined with minimal redundant networks can yield more than one edge/mapping connecting two ligands in a `LigandNetwork`

The PersesAtomMapper produces one or more atom mappings for any pair of ligands (SmallMoleculeComponents) given to it. This appears to be expected behavior, but when combined with use of the generate_minimal_redundant_network planning approach, it is currently possible that the resulting LigandNetwork features more than one edge/mapping between any two ligands. This can complicate downstream analysis of resulting AlchemicalNetworks and their results.

Is this expected behavior? It's not clear to us if this something we should be concerned about, if it's still scientifically valid, etc.

We observed this in a set of ligands used for project support within ASAP Discovery, and it broke some of our assumptions around the topology of our resulting AlchemicalNetworks.

CC: @hmacdope, @jthorton