If the Atom/RSS autodiscovery spec describes how to work with the link element to achieve feed autodiscovery in browsers and other clients, isn't it an application of (X)HTML rather than an attempt to specify (X)HTML?
My thinking was that we're accomplishing a task similar to the creators of the Robots Exclusion meta tag [1] -- put X values in element Y to achieve effect Z. 1: http://www.robotstxt.org/wc/exclusion.html#meta