On Information Gain and Regret Bounds in Gaussian Process Bandits