TurboQuant: Redefining AI efficiency with extreme compression