On the Impossibility of Convergence of Mixed Strategies with No Regret Learning