Policy Learning for Active Target Tracking over Continuous SE(3) Trajectories