"b) hypercorrection re aspiration has produced some words that managed to get accepted like μέθαύριο or εφέτος."
Why those hypercorrections—"day after tomorrow; this year", and not others? They're pretty commonplace notions, after all. LSJ has ἐφέτειος and ἐφετινός attested from ii AD, ἐφ’ ἔτος from ii BC, and καθ' ἔτος from iv BC; so it's an old hypercorrection—and I daresay not a savant one, but a product of the death-throes of when /h/ was actually still pronounced. The Latin-Greek glossaries record the "correct" μεταύριον, but I wouldn't be surprised if μεθαύριον is also that old.
So they're mis-aspirations, but i don't think they are instances of trying to resurrect old /h/ and failing, like the learnèd instances I noted in Byzantium—like John of Gaza or Paul of Aegina. They don't look like random misuses of /h/ either. μεθ' αὔριον "after tomorrow" could well be an analogy from μεθ’ ἡμέρας "with the day = during daytime"; ἔτος "year" could even have had the variant /hétos/ in classical times.
c) you correctly point out the lack of elision in new coinages, albeit savant. This is something new in that even horrible hiatuses are tolerated like your μετα-αποικιακός [post-colonial] or even τηλε-εργασία [telecommuting] or the brand new νεο-οθωμανικός [neo-Ottoman], where the unelided type is equally frequent as the elided, despite the hiatus that stands out like a sore thumb. (I assume that hiatus means χασμωδία -if not, that is what I wanted it to mean).
Yup, hiatus is indeed χασμωδία. Hiatus-avoidance is a strong feature of Greek phonology throughout its history; but we are in a transparency phase now rather than a phonological smoothing phase. That's a well-established seesaw in language change, between forms easy to break apart and understand—but harder to pronounce; and forms easier to pronounce, but harder to break apart. Hiatus-dodging is dead for prepositional prefixes in the modern vernacular: /para-/ for "overdoing something" is now just /para-/, whether it's followed by a consonant or a vowel.
Similarly, /tile-erɣasia/ *is* ugly, but it has a straightforward justification: we now only have to retain one variant of tele- in our command of Greek, and not a tel- variant. Ancient Greek had τηλ-αυγής "far-shining", and we've borrowed τηλ-αισθησία "extra-sensory perception" from the Classically-correct French télésthesie. But it's inconceivable to me that "tele-employment" (say) would now be coined as anything but τηλε-ασχόληση. And we're just not concerned about euphony any more.
In no small part, that's because of having Puristic spelling pronunciations contaminate our phonology. We're used to Ancient Greek loans and Puristic coinages sound ugly, because they violate modern (and for that matter ancient) phonotactics: we put up with monstrosities like εύθραυστος /efθrafstos/ (rather more pronouncable in antiquity as [ewtʰrawstos]). Why would be blink at /neooθomanikos/ or /tileorasi/? Learnèd words are *supposed* to sound ugly!
(And if we were Attically correct, after all, we'd never have called "television" τηλεόρασις: Attic had no time for /e.o/ hiatus either. It should have been τηλούρασις. But there's no defending that lack of morphological transparency, when /tile-/ is a dead prefix anyway.)
a) disregarding of aspiration in new compound words is not a recent phenomenon but has started more than 100 years ago, with examples like μαργαριταλιεία or the somewhat later αυτοκινητάμαξα.
It's no coincidence that Noun–Noun compounds (μαργαριταλιεία "pearl-fishing", αυτοκινητάμαξα "rail-car") dropped aspiration in learnèd Greek before Preposition-Noun compounds did. There are few prepositions and many verbs and nouns starting with /h/, so it's a rule whose application is in your face—even if you no longer pronounce /h/. It's an easy rule to remember to apply, because you see exemplars of the rule all the time.
On the other hand, how do you realise that you should aspirate Noun-Noun compounds in Ancient Greek, if you speak a Modern Greek with no /h/? Take ἅμαξα "cart": to intuitively realise that you should aspirate a noun before it, once /h/ is dead, you need to have seen another compound ending in <hamaxa>, in which the preceding noun ending in /p t k/. There are two compounds of <hamaxa> in LSJ, and one more in Kriaras; none fulfil those criteria (ἁρμάμαξα, χειράμαξα, ἀλογάμαξα). So if you're coining <autokinēt(h)-hamaxa>, you will *not* have seen any precedent to remind you to use a theta. You're relying on the letter of the aspiration rule alone. A rule that is just orthographic juggling, as far as you're concerned; so rather easy to slip up in.
So unlike prepositional compounds, you just don't see enough examples ending in <-hamaxa>—or starting with <autokinēth->—to intuit the pattern of aspiration needed. Because the rule is about written and not spoken Greek, you need to be consciously checking every compound you make in learnèd Greek, to realise where you need to aspirate. With prepositions like <apo->/<aph-> and <hupo->/<huph->, OTOH, even in written Greek, you *expect* that you'll aspirate eventually; so the check is not as onerous, and it will have a higher hit-rate. (Noone will get the rule right now, with Atticism banished; but people still had incentive to get it right last century.)
Which is why preposition-nouns and noun-noun compounds are a different story for the violation of aspiration. The prefix <aut(o)->—which Nikos highlighted in his own writing—is intermediate, because there *are* plenty of instances of aspirated <auth->: (Rattling off from LSJ: αὐθάγιος, αὐθαίμων, αὐθαίρετος, αὐθέδραστος, αὐθέκαστος, αὐθεύρετος, αὐθέψης...) So when Alexander of Aphrodisias writes αὐτοϋγιεία /auto–(h)yɡieía/ "health, in the abstract" instead of αὐθυγιεία /aut–hyɡieía/ , he really is linguistically innovating.
(Alexander of Aphrodisias, not Aristotle, is responsible for αὐτοϋγιεία btw! Alexander is citing a lost work of Aristotle, and may be modernising the aspiration.)