Attitude control for dumbbell-shaped spacecraft subject to parameter uncertainty via fully-actuated system approach and deep reinforcement learning