End-to-end RL Improves Dexterous Grasping Policies