Why reinvent this wheel? Why not instead (also) partner with people like Overleaf who already have many of these features?
Some parts of it are (duh?) but overall the system is astoundingly transparent. I think they ended up charging for the sync-to-git option, but if you look at the above repos, it appears that this feature can actually be pulled on the user end rather than just pushed from Overleaf.
Overleaf integrates nicely with existing journals and with preprint servers to some extent; I've bugged them for some time to support MarkDown (I don't care to use LaTeX for manuscripts which don't have much math) but as yet they have not implemented it. It may be an issue with the backend, although even that should be surmountable.
If a project is actually FLO, you'd think they'd make that clear to someone visiting their website.
Ultimately, it (like Paperpile) is a service that is built from mostly open components, integrates nicely with OA workflows, and has no real lock-in. I'm not sure what you're afraid of -- a paper will unsubmit itself? LaTeX will go closed source? Since your entire version history is there for the taking at any time, there's not much downside, and for our manuscripts & tech reports we've seen huge upside.
Ymmv!