Old Dog Learns New Tricks: Randomized UCB for Bandit Problems