Well, this has blown up quite a lot. I'm going to try to summarize and comment on the discussion so far, from the perspective of the "perpetrator" ;). Since we've gone off on quite a few tangents I'll send one e-mail for each one I see.
Regarding the patch itself [1]:
(tl;dr we all messed up, let's fix it)
I admittedly haven't considered its impact on extensions which might be doing surprising things to core's special pages, and I can understand how it caused issues. The severity of the issues, however, has been somewhat overstated (it doesn't require JS to create an account, like Chris said, for example).
Please stop repeating that it was "broken" (it was not; it worked as intended, you're just disagreeing with the design) and that this is not an issue in MobileFrontend (it is, if that patch is merged; MF simply becomes "incompatible" with core master). The issue of the confirmation popping up for lowercase usernames is nowhere near as insurmountable as it was painted to be and could be avoided with a few lines of JavaScript in MF (I even gave Jon a snippet on IRC).
The revert was done with an absolute total lack of communication with the people involved in the patch (there was only a brief e-mail to an unrelated mailing list[2]).
Furthermore, the blanket revert also killed the other part of the patch – live checking for invalid and taken usernames, which was a feature that has been requested for years now and which for some reason was not implemented during several reworkings and redesigns of the signup form.
While I still believe that the way this was done is the correct one (seamless warning with JS; additional confirmation step without JS), given the lack of consensus that is apparent now and wasn't apparent earlier (Steven was the only person who commented negatively on the patch) I'll rework it with a less disruptive workflow (seamless warning with JS, as before; a static short plaintext information with no confirmation step or exact warning without JS) and link the resulting patch here. Some useful comments were also left on the changeset and the bug after the revert (primarily by Jon), I'll try to address these as well (some help with the copy would be appreciated).
[1] https://gerrit.wikimedia.org/r/#/c/117234/ [2] http://lists.wikimedia.org/pipermail/mobile-l/2014-March/006557.html
(to be continued: about the deployment and about the browser testing)