Dev channel update to 82.0.446.0 is live

AmineI a lot of the theory behind rolling out features gradually is that, without a control population, you can't determine the causality of any regressions with any amount of certainty.  In other words, if we unintentionally break something when we add a new feature, especially if it's something unrelated, we stand a much better chance of pinpointing it to being caused by the addition of that feature if there are still some people without that feature that we can compare to.  Plus, having an off switch in general lets us turn things off when bad things happen, which has already proven useful a few times.