Sample-Efficient Robust Multi-Agent Reinforcement Learning in the Face of Environmental Uncertainty