That's great to hear! CodeMirror 5 has complex code for supporting bidirectional documents, but it's not perfect and lot of work. We hope to let the browser do most of this work for us in the new architecture.
Except for small drive-by contributions (see GitHub repo) the project is not in a state were we can really integrate other people in our process, and there will be pretty substantial changes to interfaces. Looking at ProseMirror, contributions will probably mostly include separate modules and not so much work on the actual core (which is pretty stable). This is especially true for CodeMirror where there will be a lot of modes, addons, themes and keybindings, which we are not planning on maintaining as part of the core project.
Improving mobile, touch and screen-reader support are from a user perspective the main reasons we are doing this in the first place. We are expecting to support most of the input behavior such platforms provide.
Yes, ProseMirror is quite stable and it is well-used. Marijn released version 1.0 nearly a year ago and since then we made some improvements, but the interface seems good enough that we didn't even discuss making a breaking change in a component as far as I know. There's not too much maintenance work necessary either, since bug reports are usually pretty good and far fewer than for example with CodeMirror. I think how happy we are with ProseMirror and maintaining it is a major motivation for making these huge changes with CodeMirror, too.
Modes implemented against the current interface will be supported by a plugin. We are definitely planning on providing a better interface for highlighting eventually, but that will take some time.
It's roughly what the two of us need (in addition to regular donations and support contracts) to be able to work on this for one year. We expect a year to be enough time for stabilizing the core and porting some essential plugins.