N translations for N epochs in deconvs instead of N-1 translations. more stable.

The only way of getting the first epoch of a deconvolution correctly aligned is shifting c_x and c_y. But c_x and c_y are constrained by all the remaining epochs and are thereby hard to move for the optimiser. Hence, I've been using N translations (no reference epoch) in my deconvolutions. It's always been stable, never had a run-away of the translation. Will try to merge main into this branch now, since we've had updates.