fix(neighbors): zero dummy cell for non-periodic systems in alchemiops naive NL

[lil-lon]

Summary

Related to #575.

Fix

The alchemiops naive backend returns a wrong neighbor list for non-periodic systems carrying a non-zero (dummy) cell, because the kernel wraps on every axis ignoring pbc.

Test

Extended test_neighbor_list_implementations with a molecule_dummy_cell axis: non-periodic molecules carrying a dummy cell must still match the ASE reference across all backends.
Confirmed that newly added tests fail without this patch.

Note

Slabs / partial pbc (e.g. [T, T, F]) are not covered and still need the upstream fix (NVIDIA/nvalchemi-toolkit-ops#104).
So, even after this PR is merged, we cannot close the above issue, but this PR would resolve the major path where molecule is called with dummy cell.

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.

[CompRhys]

Thanks! when the upstream issue is resolved will you make another PR for the slab case?

[lil-lon]

Thank you too for the review and merge!

when the upstream issue is resolved will you make another PR for the slab case?

Yes. Once the upstream issue is fixed, the kernel will handle both the slab case and the non-periodic molecule case (the case I patched in this PR) natively.
I plan to open a follow-up PR to bump the nvalchemi-toolkit-ops dependency and add unit tests covering slab-like partial-pbc systems. (Also, it's possible to drop this patch after the fix if you want.)

This PR only patched the molecule case since it was easy to fix; the slab fix is more complex, so I thought it was better to wait for the upstream fix.