Simplest possible phased-rollout implementation consuming data from the schema selected by OTA-215:

1. Nodes go into fast and stable at the start of the window for all clusters.
2. Edges go into fast at the start of the window for all clusters.
3. Edges go into stable at the end of the window (start+duration) for all clusters.

Subsequent refinements will smear out 3, but dropping all the clusters right at the end of the window is effectively what we're doing today with explicit stable-channel pull requests. And this v2 implementation would have the benefit of avoiding the VersionNotFound issue that the explicit-stable-channel-PR approach causes by delaying the node promotion (which in this v2 implementation would happen in step 1).