Informative Policy Representations in Multi-Agent Reinforcement Learning via Joint-Action Distributions