change use of `atom_tag` to `cluster_index` for internal sorting of atoms in clusters

[mastricker]

In reference to comments in PR #171

[mastricker]

Additional info: We are currently using the atom_tag, which is a unique integer, as a criterion to sort e.g. neighbourlists to minimal neighbourlists. While working currently, the safer option for us would be to use our internal numbering. The candidate information attached to an atom for this use would be cluster_index of the atoms at the lowest possible level.

Lowest possible in our case with neighbourlists would be to use the cluster indices initialized bei adaptor_neighbourlist, since the atom layer is reset (because of possible addition of atoms, i.e. ghosts)