Optimal partial questioning in large dimensional AHP