librification: Remove global state from _stbt

[wmanley]

The get/set_config mechanism is a way of storing state in configuration files. This state is global, meaning that you couldn't have different devices under test configured separately in the same process.

This commit fixes that issue by pulling a config object into the DeviceUnderTest class. This is a fairly blunt approach but it works for now. Perhaps in the future the specific keys that the DeviceUnderTest class needs can be directly injected.

I'm not 100% sold on this change yet, but I've created this PR as a place to discuss. This is obviously related to #297, although I'm not planning on ticking all the rest of the items from there off immediately.

[drothlis]

new_device_under_test needs to pass config into DeviceUnderTest.

[drothlis]

I think the config isn't the most critical bit of global state to get rid of. I'd be perfectly happy with a stbt that could be imported even if half the functions call get_config internally which reads from a global config object. I'd rather spend my time reviewing the other things on the #297 list, but if you think this is important then by all means we can proceed.

I imagine you'd also want to get rid of the _stbt.config._config global object? This pull request so far injects the _stbt.config module into the places that use get_config, but it's still a module with global state.

Whatever we do, the following must still use the match parameters from the config file:

$ PYTHONPATH=/usr/lib/stbt:$PYTHONPATH ipython
>>> import stbt
>>> stbt.match(...)

I think that's still the case with this pull request in its current form, but it's a bit hard to reason about. I think what's confusing me is the way we pass around the _stbt.config module as an object. Or maybe it's just inevitable that using dependency injection properly means you have to trace through a lot of code to figure out where things are coming from.