Towards Safe Control of Continuum Manipulator Using Shielded Multiagent Reinforcement Learning