The text itself isn't lazy-loaded – I can load any Wikipedia article, go offline, and still successfully expand all sections. Images are lazy-loaded, though, and while this has some merits, for the offline use-case this is equally somewhat annoying.

Also yes about the scroll position – while at least occasionally (possibly within the same session, but not if you're closing and later re-opening the whole browser) it seems like Wikipedia tries remembering which sections were opens and which were collapsed, it happens too late in comparison to the browser attempting the restore your scroll position. So if you had scrolled farther than you can scroll in the completely unepxanded state of the stage, your scroll position will be lost.