Sim-to-Real Transfer in Reinforcement Learning for Maneuver Control of a Variable-Pitch MAV