You're absolutely right that there are a bunch of unknowns on specifics, however, I feel pretty confident that if it's marketed as a revision, that there will be some benefit to some chunk of the legacy library, however the implementation works. The wild card would be if something like "better battery life in BC mode" was the pitch instead of better graphics, but considering that BC will almost definitely be a software solution, that seems unlikely.
Majority, no, but a lot more than I think we consider. There are hundreds (if not literal thousands) of games in the eShop developed by indies and tiny studios who have hacked together their own engines, or are running something like Game Maker and bailing wire. These are the sorts of games that 1) don't need any help from a pro model, 2) have weird stacks vulnerable to hardware changes, and 3) little-to-no dedicated QA or porting teams.
I can't imagine Nintendo making someone shell out an extra 150 bucks for hardware and the latest Musou Spin Off still runs like crap - but I can't imagine them risking breaking huge swaths of the library for no benefit. Splitting the middle, Nintendo can test a dozen or so high profile games, add them to a white list (so that they work with cartridges, right out of the box), but let devs access the new performance modes via patches, shifting the rest of the QA burden to large studios, while leaving the rest of the library functional.