Stability of Multi-Agent Learning in Competitive Networks: Delaying the Onset of Chaos