Paged, multi-request or open-ended streaming synchronization

[cdata]

As of #608 , we have a test that gives us a taste of the implications of cases when the user has not sync'd for a long time and has a lot to sync. We should be very sensitive to the possibility that the payload that needs to be sync'd is very large and contains a lot of revisions.

So, we may want revise sync to support multiple pages of updates from the client. This way no single synchronization body can ever exceed a known size.

An alternative may be to support a specialized streaming synchronization handler (kind of like the upstream version of the replication API). This API would support unbounded data transfer, while the others impose limits to the payload size they can handle.