[abhijeetgangan]

Summary

Part of #549.

• Neighbor lists fill zero-length cell vectors on non-periodic axes the way ASE's complete_cell does, so graphene(), nanotube() and similar cells keep their periodic-image pairs in every backend.
• DSFCoulombModel and D3DispersionModel return zero stress instead of NaN for zero-volume cells such as molecules, matching Ewald and MACE.
• Remove torch.jit.script from the package (normalize_inputs, math.torch_divmod).

Checklist

Before a pull request can be merged, the following items must be checked:

• Doc strings have been added in the Google docstring format.
• Run ruff on your code.
• Tests have been added for any new functionality or bug fixes.

[abhijeetgangan]

Mattersim test failure fixed in #622

[lil-lon]

LGTM. Returning zero stress for zero-volume cells makes sense. This departs from ASE, whose get_stress() raises for such cells, so it'd be good to document the convention.
Applying the same zero fill to pair potentials and Ewald / PME would make the behavior consistent.
I left the follow-ups in #549.