Policy-Gradient Algorithms Have No Guarantees of Convergence in Linear Quadratic Games