Distributed Algorithms for Learning and Cognitive Medium Access with Logarithmic Regret