per atom spin multiplicity

[chrisiacovella]

It would be good to have per-atom spin multiplicity as an option in the Loss, as this is something we can in theory compute with the multi-headed output (or potentially through modification of internals of aimnet2).