Short and Squeezed: Accelerating the Computation of Antisparse Representations with Safe Squeezing