Gravar-mail: Reinforcement learning account of network reciprocity