Random Exploration in Bayesian Optimization: Order-Optimal Regret and Computational Efficiency