Use norm of delta x within Newton scheme
Created by: chleh
The old approach of taking the residual norm was erroneous: It was computed too early (before applying Dirichlet BCs), thus it did not converge.
This PR changes that. And it switches to norm of delta x for now. That's then consistent with the way the Picard solver currently computes errors. Of course, later on there will be more error calculation methods to choose from.