Consider `ase.Atoms.constraints` for calculations.

[PicoCentauri]

ASE has an atom property called constraints omitting atom updates during optimisations and dynamics. Currently, we compute all forces even if these constraints are present. It is not problematic as the optimizer will do a correct behaviour, but we do unnecessary calculations. I think we should consider constraints when doing inference. Since we already have the selected_atoms, this should not be too difficult.

Pinging also @prajwal-pisal as he requested this.

[Luthaf]

Are you sure we can always skip force calculations? For example for FixBondLength, only the bond length is fixed, so forces on the atom might still apply a translation to the overall bond, or torque and induce a rotation.

[PicoCentauri]

Right, there we might still need them. But if there is a FixAtoms class in the list of constraints we could leave them out.

[ceriottm]

Yeah, as I commented in #212 I don't think FixAtoms can be ignored - unless there are so many that there is an atom which has only FixAtoms within its receptive radius, but then they are purely decorative and should be removed from the simulation altogether