Stop deterministic simulation if nonlinear solver failure.

Only for purely backward/forward models. In the case of purely static models the
solution for the previous or following period is not required.