Evaluating Long-Term Conversational Memory of LLM Agents