Near-Optimal Multi-Agent Learning for Safe Coverage Control