I'm afraid I don't have good news for you. In Latin one can only use meaning & context to know if the adjective/participle is used "dominantly" (NB: for a relevant terminological remark, please see TKR's comment above). Note that your first example is ambiguous between a predicative/"dominant" reading ('the highest point of heaven') and an attributive reading ('the highest heaven'). In contrast, your second example (mediis elapsus Achiviis (Virg. Aen. I, 242)) is not ambiguous for the same reason a dominant participle construction like ante Christum natum is not ambiguous either: i.e., proper names and personal pronouns (e.g., ante te cognitum (Sall. Bell. Iug. 110)) cannot be modified by attributive adjectives/participles, whereby only the dominant predicative reading is possible here.
Your third example (indeed, a very famous one!) is also very interesting: sanguine placastis ventos et virgine caesa (Virg. Aen. II, 216). Putting the relevant rethorical figure (hendiadys) aside (cf. lit. 'with blood and with a slaughetered virgin' and 'with a slaughtered virgin's blood'), I'd say that the syntactic coordination with a concrete nominal phrase like sanguine makes the attributive reading more natural (i.e., grammatically speaking). In contrast, the dominant reading is clearly preferrable in an ablative absolute context like virgine caesa, Danai ad Iliacas oras venerunt.
As far as I know, word order in Latin does not typically help to decide betweeen the two readings (althought perhaps a frequency study could give interesting hints/results, especially when dealing with lexicalized phrases). In contrast, as pointed out by Haspelmath (1987: page 31), i.a., word order is relevant in Ancient Greek: cf. his examples he hidrumene polis (attributive use: 'the city which has been founded') and he polis hidrumene (dominant use).
NB: for further discussion on so-called "dominant participle constructions" (also/typically known as ab urbe condita constructions), take a look at the following posts: here, here, here, here, and here.