Nice example btw! :)

One of the checks was erroring for me, which is what this PR fixes (maybe numpy v2 vs. v1 issue? not sure at which point this has become a hard error).

More generally, was this AI generated?

    if not isinstance(field_data, td.FieldData):
        raise TypeError("field_data must be a tidy3d.FieldData.")
    if not isinstance(permittivity_data, td.PermittivityData):
        raise TypeError("permittivity_data must be a tidy3d.PermittivityData.")
    missing = [comp for comp in ("Ex", "Ey", "Ez") if comp not in dir(field_data)]
    if missing:
        raise KeyError(f"Missing field components in field_data: {missing}")
    if (
        np.any(permittivity_data.eps_xx != permittivity_data.eps_yy)
        or np.any(permittivity_data.eps_yy != permittivity_data.eps_zz)
    ):
        raise ValueError("Permittivity_data must be isotropic.")

AI likes to do all these defensive checks. Normally, we don't really do this in our notebooks, and it would be quite verbose to do so everywhere. I feel like it can just be removed?

Regarding the isotropic permittivity check though (which is what I'm fixing), I feel like there's even a potential issue: why do we need the permittivity to be isotropic? Note that it won't be at geometry interfaces due to subpixel, which I assume is not a problem here if the recorded data does not cover the waveguide interface? But in general why enforce this at all as opposed to compute absorption components from each of the field/permittivity components separately, and add them up?