Fast Policy Learning for Linear Quadratic Control with Entropy Regularization