An approach like you described here doesn't seem very difficult to implement in general and might cover a big chunk of your use cases. It gets more difficult in case you have to edit documents that do not conform to the "normal" structure, and whenever your editors start changing the way they edit the documents in the future (without telling you of course). -- Still I agreee with Andreas that your basic workflow is broken.
What you describe reminds me of a chaotic way to edit many technical datasheets using Word, that eventually became a Plone-based system to update thousands of separate documents. My suggestion would be to rethink the way your books are being generated, since layout (including "admonitions" or "export to PDF links") is just the sauce that you add at the end of your production process.
AI might be the way to go "soon" but for now I judge it as an excellent way to break your already chaotic system because it moves any logic from your update process to the AI black box. And this black box tends to hallucinate nonsense a lot in my experience (even the newer commercial models). Maybe you could train your own custom LLM and get better results?
Maybe you could get some temporary relief from changing this
Stop accepting junk from your editors, force them to review their changes before publishing...