Parallelizing Contextual Linear Bandits