Since Mark Simonson seemed interested in the early Macintosh fonts, I thought I'd post a sample of the *original* Macintosh bitmap fonts -- i.e. the ones used on the prototype systems before Susan Kare had finalized the “city” fonts which shipped with the first macs in 1984. These are from mid to late 1983. (note that apple “restarted” its system version numbers for the final release, going from 7.x to 0.85).
One thing I’ve wondered about localized diacritics is how loan-words fit into the picture. For example, If I were to design a steeper version of the acute to serve as Polish kreska, would it be sufficient to provide localized version of only [ćńóśź]? If, for example, my name were to occur in a Polish text, would Poles find the acute accent incongruous because it failed to match the kreska?
This raises the possibility, though, of a potentially huge proliferation of localized variants (unless, of course, one were to decompose all accented characters in the ccmp feature and to provide localized versions of combining marks only).
I'm sure that plays a role, but the other thing to consider is contrast. For example, the modern greek tonos is often near-vertical which would not have been acceptable in polytonic greek where the oxeîa (acute) and vareîa (grave) needed to be kept distinct. For similar reasons, a steep kreska in polish is fine but I suspect that a very short kreska would be much less acceptable than it would be in a language which does not need to keep it distinct from the dotaccent.
P.S. You can easily refer to the beginning of a paragraph, just not to the character itself or anything from the previous paragraph.
ignore sub @AllCharacters A; sub A' by A.init;
can be used for a paragraph-initial alternate, but it won't help with the problem you're trying to solve.
(I bring this up only because I just spent some time fixing a font that used sub space X' by X.init for all of its word initial alternates which, of course, prevented them from being used at the beginning of a paragraph).