[...] abandoned by Tolkien?
Yes, PE17/23 gives "√GIL √ÑGIL". But I only shared that information to say that even in proper compounds a prenasalised G that follows a liquid would have stayed as it is (just as the D of dor remains in Mordor, instead of becoming Ð/DH, because it came from NDOR). Not that it actually relates to the "false compound" situation.
I call them "false compounds", because if they were regular compounds like elvellon, then they would have had aur as a following element (like a suffix "-or"), as in *Menelor "sky-day = day of sky". Because "Ormenel" does not make sense (as a word for a weekday) as "day-sky = sky of day", hence it is rather "day (of) sky", which is what Sindarin normally does with phrases (e.g. Aran Moria "King (of) Moria"), but almost never with compounds. Almost, hence these "or-" words.
Not just that the word became polysyllabic, but the fact that AU came to precede a consonant cluster would have "forced" it to become O as most AU from ancient AW (cf. glaur > Glorfindel). The name Aerandir does the opposite for the same purpose (of lessening syllable weight) where the double R is shortened to single as it follows a diphthong. Though there is also Orithil (which lacks a consonant cluster), but at this point I can play the "reformed by analogy" card. As for which rules apply and which do not, our sample size is too small to tell.