fix(lammps): narrow bond_expansion type for mypy

nn.Module's __getattr__ types attribute access as Tensor | Module; the
else branch assigning ``be`` directly to ``self.bond_expansion: nn.Module``
tripped --strict mypy. Cast at the point of access (same pattern used for
potential.data_mean elsewhere in this file).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: shyuep

src/matgl/ext/_lammps.py

@@ -134,7 +134,11 @@ def __init__(self, model: nn.Module) -> None:
         # the kernel survives torch.jit.script.save. Gaussian / ExpNormal /
         # RadialBessel basis modules are already TorchScript-friendly and pass
         # through unchanged.
-        be = model.bond_expansion
+        # ``model.bond_expansion`` typed as ``Tensor | Module`` via nn.Module's
+        # ``__getattr__``; the cast narrows it for mypy without runtime cost.
+        from typing import cast
+
+        be = cast("nn.Module", model.bond_expansion)
         if getattr(be, "rbf_type", None) == "SphericalBessel":
             if not bool(be.rbf.smooth):
                 raise NotImplementedError(