Ability to calculate a viable estimate of replication completion

A side effect of the efficiency savies we get from parallelising Pull replication is that the stats you get from the replicator are difficult to use to show a progress bar. This is because as you complete replication for more documents, more are being added to the total. It depends a little on latency etc, but this generally ends up looking like you're in the 85-95% range for the whole replication.

I realise that this is quite closely tied to the implementation, but I wanted to ensure we have a ticket to track this.