Learning Competitive Equilibria in Exchange Economies with Bandit Feedback