Andrew/refactor

[ayjayt]

This is a relatively large refactor but each commit is well written in stages.

That said, intermediate stages (commits) don't function.

---

description

We start with creating browser specific classes that ONLY return the CLI and env vars for a specific brand.

This lets us split the stages:

1. processing starting/stopping in its own idea
2. its own classes generating proper cli/env vars (must read the Pipe, now called channel, see below)
3. a separate browser_wrapper now with above functions removed so the wrapper does it's one function of
   securing the proper file descriptors.

This requires two other refactors:

First:
Pipe will be refactored to have a standard interface (although it wont necessarily be enforced)
Pipe will split out serializer so that other channel types (like WebSocket) can use it.
Second:
System utilities needs to better divide its concerns:

1. finding paths
2. downloading a browser
3. getting a specific path
4. temp files

Then, the last refactor will be splitting browser into sync and async, where async will be composed with a broker class to separate the ideas of process control and async manage (sync doesnt really need a broker since sync can't do any sort of management anyway since it can't deal with the inherent asyncronicity of the process-process interaction).

After that, tests will be fixed since namespaces will have changed, even if expectations havent.

Documentation will then be included and built.

And we launch v1.0.0rc on testpypi.