TransientPropagation: Always Extend Induction Matrix

If we have non-optimal weighting potential maps that are not zero at their borders, we get fake induction whenever we cross a pixel boundary because charges would always move into the weighting potential map (creating a positive d\phi) but never out of the map.

We can correct for this to some extend by extending the induction matrix artificially, similarly to what we have done already in !435 (merged). Now, we extend always, in order to also cover the move out of the matrix.

Note that this does not happen if a proper weighting potential is used, where the field really falls to zero at the border.

Might be a candidate for backporting to v1.6 - if we want to do this at all.