Add computations with cell and neighbor list to the DirectPotential class

[PicoCentauri]

As in the title, I added the ability to pass a neighbor list and a cell to the DirectPotential class. A user can only pass a cell and neighbor shifts to take the periodic boundary conditions into account and/or additional neighbor indices to limit the pairs for which the potential should be computed.

For the background. I need this extension to subtract "intra" molecular interaction between atoms. To compute these it is mandatory to pass explicit neighbor indices and the cell.

While adding this feature I unified the distances operations and put them together with a function for computing all possible neighbor indices of a system into a file called neighbors.py. I am not super happy with the file name. If there are better ideas, I am happy to rename.

I also fixed some layout issues for the potentials.py and enabled the rendering in the docs for this file.

TODO

• Test DirectPotentail with neighbor_indices
• Test DirectPotentail with cell and neighbor_shifts
• Test DirectPotentail with neighbor_indices, cell and neighbor_shifts

[E-Rum]

LGTM!

[E-Rum]

Why do we prefer to use a relative import path (..lib) instead of an absolute path (torchpme.lib)? The absolute path looks cleaner to me :)

[PicoCentauri]

Interesting for me other way looks cleaner. We/I prefer the relative imports (within a package) because it may make refactoring/renaming easier. If the folder structure changes you don't have to touch anything. Same for renaming.

Also, imports are shorter. You don't have to write from mypackage.XXX.YYY.ZZZ import myfunction if myfunction in the same directory just from .ZZZ import myfunction.

On top there is a style issue. isort will split the package imports into the groups seperated by an empty line:
('FUTURE', 'STDLIB', 'THIRDPARTY', 'FIRSTPARTY', 'LOCALFOLDER'). which for us will result in:

from typing import List, Optional, Union

import torch

from ..lib import InversePowerLawPotential, all_neighbor_indices, distances
from .base import CalculatorBaseTorch

if we would use absolute imports it might not be super clear what is THIRDPARTY and what is FIRSTPARTY

from typing import List, Optional, Union

import torch

from torchpme.calculators.base import CalculatorBaseTorch
from torchpme.lib import InversePowerLawPotential, all_neighbor_indices, distances

Maybe this is convincing.