Tight Bounds for Collaborative PAC Learning via Multiplicative Weights