The Impact of Batch Learning in Stochastic Linear Bandits