Bridging the Theoretical Gap in Randomized Smoothing