Autonomous Sailing With Sim-To-Real Reinforcement Learning