The Dormant Neuron Phenomenon in Multi-Agent Reinforcement Learning Value Factorization