Self-consistent loop for non-linear controls

[goerz]

When the Hamiltonian is non-linear in the controls, i.e. ∂H/∂ϵ(t) still depends on ϵ(t), there is an issue where the updated field occurs on the right-hand-side of the update equation. The default approach to handle this has always been to use to approximate the unknown updated field with the guess field. This is only valid for small pulse updated (large λₐ).

It should be relatively straightforward to add a self-consistent loop here, where in the first iteration, the update is calculated with the guess field on the right-hand side, and then the pulse update is re-calculated with that update. The loop continues until the update reaches a fix point, up to some desired precision.

This should allow, e.g., pulse parameterization with smaller values of λₐ, and thus higher convergence.