Learning Two-agent Motion Planning Strategies from Generalized Nash Equilibrium for Model Predictive Control