From Tools to Teammates: Evaluating LLMs in Multi-Session Coding Interactions