The problem with MS Word HTML is that wherever you paste it, it’s sh t in, sh t out unless the HTML is cleaned up by the editor (which as you pointed out is tedious). Word unfortunately does not create great markup, so you either leave the HTML as is (making it harder to post-process) or clean it up.
Once you cleaned it up, it does not matter whether you used a well-defined ProseMirror JSON schema or clean HTML - those two should be equivalent at this point.
We ended up writing a Word importer that detects elements and transforms them into a clean document that ProseMirror can display. Pandoc for instance does a reasonably good job in doing so.
Ask yourself: How important is a really well-defined document (maybe even with semantic structures for you). If it’s not important, an editor that handles any HTML is fine. Otherwise, you will need to do some work to make it fit your use-case.