The potential performance improvement may come from cythonised dict<int> lookup
vs pure pythong function call of the Hamiltonian value.

Further performance improvement can be made by making tinyarray cimport-able
and use libcpp.map instead of dict when saving static onsite values.