Add support for advanced OpenFF handler converters

[SimonBoothroyd]

Description

@aehogan this is roughly what I was thinking of. Basically it means you can take the already converted electrostatics potential that contains all charges (including due to v-sites, library charges, AM1BCC etc), and then create a new potential that includes both these plus the info from the multiple handler along the lines of

@smee.converters.smirnoff_parameter_converter(
    "Multipole",
    {
        "polarity": _ANGSTROM**3,
        "cutoff": _ANGSTROM
    },
    depends_on=["Electrostatics"]
)
def convert_multipole(
    handlers: list[
        "smirnoff_plugins.collections.nonbonded.SMIRNOFFMultipoleCollection"
    ],
    topologies: list[openff.toolkit.Topology],
    v_site_maps: list[smee.VSiteMap | None],
    dependencies: dict[str, tuple[smee.TensorPotential, list[smee.NonbondedParameterMap]]]
) -> tuple[smee.TensorPotential, list[smee.NonbondedParameterMap]]:
    import smee.potentials.nonbonded

    (
        potential,
        parameter_maps,
    ) = smee.converters.openff.nonbonded.convert_nonbonded_handlers(
        handlers,
        "Multipole",
        topologies,
        v_site_maps,
        ("polarity"),
        ("cutoff"),
    )
    potential.type = smee.PotentialType.POLARIZATION
    potential.fn = smee.EnergyFn.POLARIZATION

    electrostatics_potential, electrostatics_potential_maps = dependencies["Electrostatics"]
    # TODO: merge the two handlers

    # just returning the new potential will lead to the original electrostatics potential being removed,
    # so this new potential should handle all electrostatics interactions.
    return potential, parameter_maps

Status

• Ready to go

[SimonBoothroyd]

the fixed charge potential (either in vacuum or with PME) can be computed using compute_coulomb_energy, and we should be able to just autodiff w.r.t. the coords so we should be able to get the field also in compute_multipole_energy.

@aehogan if this approach looks reasonable, i'll finish off this PR and can also give a better example of what convert_multipole would look like

[aehogan]

This looks great and increases flexibility for more complicated potential forms in the future.