Transformer-based Value Function Decomposition for Cooperative Multi-agent Reinforcement Learning in StarCraft