Optimal subsampling for large‐sample quantile regression with massive data