Mark adds via e-mail (paragraphs are from two separate messages):
I’m pretty sure some of the New Guinea data I’m looking at shows segmental nasals ONLY appearing in coda positions, phonologically, though due to nasal spreading from nasal vowels they are heard in onsets as well, if there’s no other recourse.
To elaborate briefly on the nasals: in Damal there are three phonetic nasals, predictably [m], [n] and [ŋ]. [ŋ] only ever appears as a coda; [m] and [n] do their best to appear in coda positions always (including VNV sequences), but can be found in onsets if they have to be (#NV# sequences are found), but there’s strong evidence that these are underlying /D/ + nasalised vowel).
I wrote back:
Speaking of nasals: in all varieties of Spanish, there are three phonemic nasals (bilabial /m/, alveolar /n/, and palatal /ɲ/), as determined by contrasts that surface word-internally before vowels (i.e., in onset position). A nasal is neutralized syllable-finally: it either assimilates to a following consonant (resulting in one of seven phonetic nasals) or it surfaces as a default nasal word-finally. In many varieties (including the standard), the default is alveolar [n]. In others, the default is described as velar [ŋ]. In these latter varieties, what’s interesting is that the velar nasal is not one of the phonemic nasals.
I have a paper on this that cites much of the relevant literature. It includes a theory-internal argument for the position (originally taken by Trigo 1988) that the so-called velar nasal is actually a placeless nasal, which may or may not be relevant to your concerns.