Check for plausible likelihoods in case of nested models

[dweindl]

Maybe I missed it, but I don't think it's implemented yet:
It would be nice (to have an option) to automatically check whether in case of nested models and identical datasets, the super-model has a likelihood at least as good as any submodel. If this is not the case, this is most likely due to non-converged optimizations, and should be communicated to the user.