Signature for legacy toolkit atom typing and NN atom typing.

[yuanqing-wang]

Related to discussions we have here #9

I suggested that the signature for the function for legacy atom typing frameworks as well as NN atom typing / parametrization should be the same, i.e. graph -> graph.

Nonetheless, this would mean that we'd have to carry the molecule (RDKit, OpenEye, or OpenForceField) object with the homogeneous graph. This wouldn't be very efficient.

If we have the function for legacy typing to have the signature molecule object -> graph, we need to think more about the bookkeeping efforts to make sure that the indices won't be messed up after batching, training, and debatching.