Hate to rain on the parade, but I have a few points to make (I haven’t read the paper and I’m not a C++ mastermind, so who knows what crazy stuff they talk about in there):
First of all (and this I’m sure we all know), Epic isn’t going to add features to C++ 17, so if that was a semi-serious idea it’s probably best to let that one go right away. If Headerless code is even possible in C++ for a code base as large as this, then until some huge company does it like MSDN or something and makes it the Standard (and they throw out all the rulebooks on C++ and start again), then it’s not gonna happen.
Secondly, I don’t even think headerless code is possible. Unless compilers / linkers change as well, then surely all we’ll end up with is bloated binaries - since each class you create would have to ‘include’ every other class, and somehow avoid circular dependencies. Just because you can doesn’t mean you should…
And of course the main thing is, any effect this has on compile / iteration times will more than likely be negative, not positive. You won’t be able to just compile .cpp files anymore for fast iteration, you’ll have to recompile the entire code base - which is just ridiculous.
TL;DR: Unlikely to happen IMO. A quick google search indicates why Headerless C++ (if possible) would be a bad idea.