Break up the driver into smaller pieces

The current implementation provides a _monolithic_ driver where the AST and the communication with the server are in the same module.

I feel that the code responsible for talking to the server should not be bundled with the AST code (and maybe other separations should be added too, but I've not digged into the various concerns to produce helpful ideas here).

In our software, we have a _general querying layer_ which produces ReQL expressions. These are then serialized and passed to a querying engine, which runs them agains the RethinkDB server. In this scenario, the _general querying layer_ needs to include the driver as a dependency in order to create the ReQL expressions. And this looks like a code smell. Only the querying engine, which needs to talk the the server, should require the driver. The higher level abstraction should be able to only use the AST portion of the library.

What do you think? Am I overly concerned ;-) about this separation of concerns?

Start by mapping the current driver module and identifying which parts build ReQL AST expressions versus communicate with the RethinkDB server. Review how expressions are serialized and passed to the querying engine. Done means the AST portion can be used independently by a general querying layer, while only the querying engine depends on server communication.

Written by the indexing model from the issue text.