Multi-agent DQN with sample-efficient updates for large inter-slice orchestration problems