Scaling Up Differentially Private LASSO Regularized Logistic Regression via Faster Frank-Wolfe Iterations