The Whole is Better than the Sum: Using Aggregated Demonstrations in In-Context Learning for Sequential Recommendation