Pure Exploration and Regret Minimization in Matching Bandits