Fast and accurate convolution-type smoothed quantile regression. Implemented using Barzilai-Borwein gradient descent with a Huber regression warm start. Construct confidence intervals for regression coefficients using multiplier bootstrap.