Process for deciding a feature is good enough to make non-experimental

Ib !2569 (merged), @wesleyac writes:

(I am assuming here that we want to make the memquota feature non-experimental, I haven't been following this closely enough to know the status of that. We should perhaps document the process by which we add new feature and make them non-experimental/default.)

I quite agree. I'm not sure what this ought to look like. When there were just three of us, "the MR and the MR reviewer agree" was clearly sufficient - and people would typically be more aware of other stuff going on.

Now the team is bigger we probably need a wider set of opinions, and a more formal way of soliciting them. We also need a way to decide what level of review/control ought to be applied (at least, some guidelines that an MR reviewer could use, I guess).