Near-Optimal No-Regret Learning Dynamics for General Convex Games