Multi-Agent Learning in Contextual Games under Unknown Constraints