Has anyone done this before or does anyone have any tips on how to populate a ve.dm.DocumentNode from HTML not in the VisualEditor?

The VE data model (with ve.dm.DocumentNode at the root) is designed for our specific representation of documents in HTML. You can convert any HTML document into VE DM, but it is designed to extract and make editable the information that VE considers important (for example, text colour is not easily accessible in the model).

The closest we have to comparing two HTML documents is the historical diff viewer, which uses ve.init.mw.DiffLoader to load two revisions from Parsoid. It requires most of the VE codebase (and extensions) to parse the document, and render the diff. (Sepcifically ve.dm.Converter, and all the ve.dm.*Nodes/Annotations which the converter calls out to).

The example you link to is the closest we have to a demo of tree diffing outside of VE right now, and most of our future efforts are going to be focused on the VE integration, but we're always interested to see downstream uses of our libraries!

Wow, that's a big compliment! Thank you! Unfortunately, this is a lot of work, and it's not something we're going to get around to doing for a very long time.

That's understandable. I'm using CKEditor 5 in my app, and they have their own data model and virtual DOM. A standalone tool does seem the best for a generic solution. Almost every diff tool out there assumes the content you're diffing is strings but the need is certainly out there to diff HTML. Daisy diff is difficult to integrate and very old, and other packages like nokogiri-diff are broken out of the box. I'll come back and comment if I develop my own solution.

Cheers!