Compose models for allegro-pol simulation

[CompRhys]

Adds SerialSumModel to allow for allegro-pol to make predictions and pass to subsequent model in single sum model.

[orionarcher]

Not sure I fully understand the purpose of UniformPolarizationModel or when it should be used, which probably betrays either my ignorance or the need for slightly more documentation. Implementation seems good aside from a few docs and interface suggestions.

[CompRhys]

Not sure I fully understand the purpose of UniformPolarizationModel or when it should be used, which probably betrays either my ignorance or the need for slightly more documentation. Implementation seems good aside from a few docs and interface suggestions.

So the aim of the contribution is that if you run a simulation of polarizable material in an electric field the additional terms are just a sum contribution to the overall forces and energy in the small field limit that models like allegro-pol were trained for.

The challenge was that with the first SumModel I added last week the can't handle the case where the first model generates extras that are used by the second. The serial sum model here allows for that [h(x) = f(x) + g(x)) vs h(x) = p(x) + q(x, r(x))]. So if we call allegro-pol to get the polarization and born charges then we can take those as the input to the UniformPolarizationModel and get the energy contributions from the uniform electric field.

The UniformPolarizationModel name comes from the fact that we should also have a SpatialPolarizationModel which would let you vary the E field in the cell. The maths there is more involved because of image charges and PBC. That can be a later addition.