[Development] Add a divergence tolerance for Newton's method

[sblauth]

At the moment, it may happen that Newton's method diverges and produces larger and larger residuals. Users should be able to specify some (relative) divergence tolerance, so that when this tolerance is surpassed, the Newton solver fails.

For optimization this has the benefit, that less iterations are needed to compute when Newton's method diverges and a smaller stepsize can be used faster.