Note, there are Scrivener users who successfully use Zotero without issue. There were two workflows before Zotero 7 was released, and the ODT scan route you used was removed with the new Zotero version (though a motivated developer could probbly fix it I think, just that the original dev has moved on, this is part and parcel of free and open source). But there is still a workflow that, in my opinion, is superior, that uses temporary citations like you used before, but replaces ODT scan plugin with Zotero’s amazing BetterBibTeX plugin allowing you to use citekeys and thus any output compatible with that.
The wiki post covers the details, but Scrivener supports this workflow explicitly, it was designed for temp citations like this and automated output to tools that will build your final document, including (not limited to) things like LaTeX that are leagues ahead of Word for academic content.
EDIT: I do overall agree that L&L should think of better ways to integrate bibliography generation into their compiler (the BBT workflow has several moving parts and does require effort to set up, it is not plug-n-play). My suggestion Academic Bibliographies: support citeproc as an option during compile has been for them to keep the temp citation workflow as-is, but add a compile-time option to use a citeproc engine (like Zotero, pandoc, mendeley and many others). This would work for any reference manager, and is relatively simple (citeproc is designed with an API to replace temp with final cites) allowing Scrivener to offer native bibliographies in any of its output formats. Replacing MMD with Pandoc would also allow them to offer this for any of their MMD-based outputs.