Enhancing Trust-Region Bayesian Optimization via Newton Methods