Distributed Area Coverage with High Altitude Balloons Using Multi-Agent Reinforcement Learning