sparse solver shouldn't need to make a supercell

[ExpHP]

The reason why Phonopy needs to make a supercell is because it takes atomic forces as input for each displacement. However, rsp2 now has the concept of bond forces.

Force constants calculation should use bond forces, and there could be a helper type for using a supercell to obtain bond forces from a potential that only provides atomic forces.

[ExpHP]

Triage: Currently irrelevant because rsp2 never even computes phonons away from the gamma point anymore.

[ExpHP]

Addendum: IIRC, I also later discovered that a supercell is still, in fact, necessary, even when using bond forces, but I don't recall why. (I was hoping to find that explanation here...)

---

Edit: I think I might remember what it is, and it basically just boils down to the following:

• There's that thing mentioned in the config doc comments about an atom interacting with images of itself. (e.g. in graphene unit cell, moving an atom along the bond would change the bond angle because the atoms two neighbors away also move)
• Not all potentials support bond forces and thus still require large supercells (e.g. lammps). So updating the force constants code to support bond forces would increase complexity for very little gain.