ASALT: Adaptive State Alignment for Lateral Transfer in Multi-agent Reinforcement Learning