Attack and defense against deep reinforcement learning-based agent for discrete domains: Application in distribution network reconfiguration