RVT-2: Learning Precise Manipulation from Few Demonstrations