Improve accuracy of PBC computations

[WardLT]

We see some differences in the forces acting on atoms that are symmetrically identical. Small, but enough to worry me.

See #28

[WardLT]

The problem is likely that our use of a fixed number of neighbors per atom. With high symmetry, there will be many atoms at the same distance in the neighbor shell and - due to the annoyances of numerics - atoms with the "same" structure are getting different subsets of atoms in a certain neighbor shell. Atoms at different points (but the same distance form the center) will lead to different energies and, especially, different forces (imbalance around the origin)

[WardLT]

I'm debating making this a #wontfix, as it only harmful on edge cases and holding a fixed number of neighbors is a choice we made for speed (fixed array sizes are faster on GPUs)