Add "rollback" RTV to experiments.html for bisecting regressions

Often when something breaks, the first step is opting into the previous Stable RTV to see if the error can be reproduced. It could be useful to have an explicit "Previous Stable Channel" next to the Experimental/Beta opt-in channels so the release on-duty and others can quickly verify if a release or cherry-pick introduced the breakage, or if it was a long-standing issue.

/cc @ampproject/wg-infra