[NEW] Inject slot migration states into RDB

[PingXie]

          I know we discussed it, but I forgot the discussion. I still think it's odd that we are putting the replica online before it has the slot information, since it will have a point in time but won't serve any data. It also is getting added to the replication buffer, which could cause replicas to unnecessarily be unable to psync. My previous PoV was why not inject this into the RDB file, which is the true point in time which we are trying to stick on at the end. We already stick "extra" metadata into the RDB file like the psync offset. If a replica comes up, it will also have a partial state.

Originally posted by @madolson in #245 (comment)