To Compress or Not? Pushing the Frontier of Lossless GenAI Model Weights Compression with Exponent Concentration