I don't know that there is any good reason to allow a move from visual mode to ex mode (or vice-versa). I do think that ex mode is very important - though I would expect you could implement it as its own UI over msgpack?
As neovim matures I think the opportunities for rich extension and scripting are going to blow away anything vi/vim users have ever had. I think ex-mode could be reimplemented as you suggest, or we'll end up with other awesome approaches that were simply unthinkable in the old architecture and codebase.
It looks like tarruda explicitly says, "As for the ex command-line utility, that can easily be implemented as separate program that talks to nvim via msgpack-rpc".
This seems like a reasonable separation of concerns.