[WIP] HTTP/1.1 - #92

Alright, I think I have completed all my initial tickboxes.

I want to write some release tests before I merge this, but I'm beginning to think this is nearly done. There are still some outstanding work items for cleaning up:

• We reach into hyper.http20 for a few things (exceptions etc). These should be moved to hyper.common instead.
• Some of the methods aren't quite named the same (getresponse() vs get_response()). For now, let's move to the slightly better names on the HTTP/1.1 classes.
• HTTP/1.1 responses have a few extra methods that I should really support on HTTP/2 for the purposes of the abstraction layer. The obvious one here is read_chunked, which should at least be easy to implement for H2.
• Need some documentation!

These failures are difficulties with rabbitmq's APT repository, which we only hit when we try to build nghttp2. They're intermittent, and annoying.

Sadly, at this point I need to reconcile the different header handling logic between the HTTP/2 layer and the HTTP/1.1 layer.

This turns out to be really painful, because the HTTP/2 layer deals with headers as unicode strings and the HTTP/1.1 layer currently does so with bytestrings. I think I can make this work by adding a unicode interface to the HTTPHeaderMap, but it's really painful. =(

Ok, I think the series of commits I just put up solves that problem. It simultaneously cleans and dirties the API, which is weird.

@sigmavirus24 @shazow Is there any advantage in merging-and-releasing this block of code before I put a compatibility layer in front of it?

No dis-advantage. :)

You're infinitely more likely to never release it if you wait, vs if you release.

Without the compatibility layer, people might start testing it and using it so you get feedback sooner. That's the only advantage I can think of.

Ok, screw it, I'm going to merge this. I'll make a release that contains only this change on top of v0.2.2 (which I've just released), so people will always have a previous-stable-version to move to if for some reason this breaks them.

Deep breath people, let's merge this 71-commit monster!