A Bayesian Optimization Approach to Compute the Nash Equilibria of Potential Games using Bandit Feedback