Collaborative Learning and Personalization in Multi-Agent Stochastic Linear Bandits