Deep Reinforcement Learning for System-on-Chip: Myths and Realities