The way that CCS works "under the hood" is something of an implementation detail. As it stands today, remote cluster connections use the so-called transport protocol, just like connections between nodes in a single cluster. This means that Elasticsearch opens multiple parallel TCP connections and keeps them open for an extended period of time even when they're not in use. However there's no real guarantee this'll be true in all future versions.
Not really. The internals are an implementation detail, so there isn't a lot to document outside of the code and its tests. If you have a more specific question, perhaps about how to set up your network for CCS, then perhaps we can help with that.