Document cross-datacenter latency issues

[coffeemug]

The latency implications of our query model (i.e. masters, automatically hopping to masters if you're querying replicas, and the outdated flag) are not obvious to users (for example, see https://github.com/rethinkdb/rethinkdb/issues/2115).

We should write a document that describes best practices for multi-datacenter setups and goes into detail on latency implications.