perform _pre_replicate inside xdcr

As of now, xdcr makes one _pre_replicate call for each vbucket when replication starts up. Each pre_replicate call is a remote call to the target cluster. This can be made more efficient by having xdcr make one remote call to retrieve the failover logs on target, and perform ___prereplicate inside xdcr using the failover logs. This will also remove one dependency of xdcr on ns_server, which is the current handler of the __pre_replicate calls.