Semi-Lagrangian schemes for mean field game models