Replication breaks if WAL is missing

[Kostja Osipov]

If we start a replica, get some updates, then stop it, then save a snapshot on master, then purge some old (pre-snapshot) logs, replication breaks.

We need to be able to figure this out on the replica and use the data in the snapshot instead.
This would also simplify replica configuration -- it would be sufficient to point a replica at a master, and it will automatically fetch all necessary data. Today one has to copy over the initial snapshot by hand.