On the interplay between exposure misclassification and informative cluster size