[jhauschild]

Motivation is to address #164 , which is essentially the part 2 inheritance of #254 for #255.

The second commit here does this, making the DegreeOfFreedom attributes of the Site instead of subclassing them.

Downstream: Site DOF composition (lasyten) → TeNPy v2_main

Follow-up for the lasyten Site/DOF composition refactor (issues #164 / #254; call-site map:
convert_SiteDOF.md).
After #611, TeNPy models already go through lasyten Sites + Coupling factories. That path is mostly unchanged. What breaks is anything that still assumes DOF mixin
inheritance (SpinSite is-a SpinDOF, flat attrs on the site, isinstance(site, FermionicDOF), etc.).

What changed in lasyten (Python)

Old	New
SpinSite / SpinHalfFermionSite / … subclass DOF mixins	Concrete sites subclass Site only
site.spin_vector	site.spin.spin_vector (or site.dofs["spin"].spin_vector)
site.Nmax / n_tot / JW / number_operators / …	site.occupation.<attr>
site.clock_operators	site.clock.clock_operators
isinstance(site, SpinDOF) / FermionicDOF / …	site.spin is not None / isinstance(site.occupation, FermionicDOF) / …
site.creation_annihilation_ops(...) (via MI)	lasy.FermionicDOF.creation_annihilation_ops(...) (static on the DOF type)
site.num_species on every occupation site	Still on SpinlessFermionSite; for SpinHalfFermionSite use site.occupation.num_species
Conventional DOF registry keys: "spin", "occupation", "clock", "anyon".

Does not need changes (if used as today)

• Site constructors (SpinSite(...), SpinlessBosonSite(Nmax=...), SpinHalfFermionSite(...), …)
• Named onsite ops via site.get_op(...) / add_onsite_operator / model add_onsite / add_*_coupling with op names
• lasyten coupling factories (spin_spin_coupling, hopping, clock_clock_coupling, Hubbard-style factories, …) — they take Sites and resolve DOFs internally
• isinstance(x, lasy.Site) / concrete site classes (SpinSite, SpinHalfFermionSite, …)

Required TeNPy edits on v2_main (checked against current tree)

1. tenpy/models/spins.py — only production hit

• _onsite_matrix: site.spin_vector → site.spin.spin_vector
• _coupling_matrix: site1.spin_vector / site2.spin_vector → site1.spin.spin_vector / …
• DipolarSpinChain.init_couplings: s.spin_vector → s.spin.spin_vector
  2. tests/test_model.py — test_model_plus_hc
• s.spin_vector → s.spin.spin_vector
• s.creation_annihilation_ops(s.num_species) → e.g.
  lasy.FermionicDOF.creation_annihilation_ops(s.occupation.num_species)
  (and use s.occupation.num_species for SpinHalfFermionSite; SpinlessFermionSite.num_species still exists on the site)

Explicitly not broken by this refactor (on v2_main)

• Other ported models from #611 (tf_ising, clock, fermions_spinless, hubbard, haldane, hofstadter, …): they construct sites and define couplings by name / factories — no flat DOF attribute access found.
• Passing Nmax= / conserve= / filling= into site constructors.
• Docs/examples that only mention site class names (no flat DOF attrs).

Suggested smoke checklist after bumping lasyten

• SpinModel / SpinChain / DipolarSpinChain construct and build H
• test_model.py::test_model_plus_hc for spin, fermion, spin-fermion
• Optional: SpinHalfFermionSite model (e.g. Hubbard) still builds — should already, but confirms coupling factories