On the Robustness of Global Feature Effect Explanations