Computational Aspects of Bayesian Persuasion under Approximate Best Response