Learning in Congestion Games with Bandit Feedback