Add selector indexing of AbstractHamiltonians

[pablosanjose]

With #277 we can now do h[sites(...), sites(...)] with h::AbstractHamiltonian. This gives a SparseMatrixCSC, a copy of the corresponding matrix elements of hamiltonian(h). It parallels g[sites(...), sites(...)] with g::GreenSolution, which gives a Matrix. But with g we can also do g[::SiteSelector, ::SiteSelector], which gives an OrbitalSliceMatrix. For consistency, we should also be able to do that with h, i.e. h[::SiteSelector, ::SiteSelector], also returning an OrbitalSliceMatrix

(Incidentally, h[sites(...), sites(...)] should return the corresponding block of default_hamiltonian(h), not of hamiltonian(h).)