Bridging Temporal Gaps: A Distribution Shift-Aware Approach for Long-term Data Utilization in Short-term Traffic Prediction