A semi-Lagrangian algorithm in policy space for hybrid optimal control problems