The particular linguistic hypothesis to which you refer posits that nasal vowels are the result of an underlying /VN/ (where V is any vowel and N any nasal consonant, not just /n/) in the same syllable.

So plein in this analysis is underlyingly //plɛn// (one syllable) while pleine is //plɛ.nə// (two syllables). The nasalition process then affects the syllable //plɛn// but not //plɛ//, yielding /plɛ̃/ and /plɛ(ː)n(ə)/.

Applying this analysis to onde then, we’d start with an underlying //ɔn.də//, yielding //ɔ̃.də// after the nasaltion rule, then /ɔ̃d(ə)/ once the optional schwa deletion rule is applied.

I’ll note that this analysis, like all hypotheses who try to posit that a diachronic sound change has become a synchronic phonological operation, is really vulnerable to exceptions to the sound change and to later changes or borrowings. Even ignoring newer loanwords, one issue are words such as hymne /imn/ (that was /i.nə/ at the time nasalisation happened, with the /m/ inserted later by reading pronunciation), that this hypothesis would predict should be pronounced /ɛ̃n/ (from an underlying //im.nə//). Another are pairs such as ennui /ɑ̃.nɥi ~ ɑ̃.nwi/ and innoui /in.nu.i/ or /i.nu.i). If you posit that the underlying form of ennui is /an.nʏ̯i/ (as you’d have to do to explain its nasal vowel by this hypothesis), then the absence of nasalisation of the geminate form of innoui becomes puzzling.