We needed to implement `GraphState.from_json` so that the server can
track a copy of the client's state. That's a useful byproduct of this
work!

Moved that functionality into `abp.GraphState`, so that you can
just run `state = abp.GraphState(nx.Graph(((0,1),)))`, `state =
abp.fancy.GraphState(nx.Graph(((0,1),))` or whatever.

Now you can write `abp.fancy.GraphState(nx.Graph())`.
TODO: make this work for a non-fancy `GraphState`

Trying to get more sensible defaults for people who like to think about
graphstates rather than circuits.

Repsponse to issue #4: [Link](https://github.com/peteshadbolt/abp/issues/4)

I have written a tonne of tests that compare against A&B's original C++.
Until I can distribute the `anders_briegel` package which implements
that original code, we now skip those tests if we see an ImportError.

TODO: seems a little over-zealous right now ...

- `GraphState.copy()`
- `measure_x` etc.
- `measure(detail=True)`
- Better documentation on `remove_vop` and `act_local_rotation`

Still need to "test the tests", but it's looking good.

Next up is just to clean those tests and replace the stupid
act_local_rotation2 with something legible.

Looks like nondeterminism in `next(d.iterkeys())` was breaking
consistency with A&B. Next up: switching determinism on or off.

Had a dumb translation error

Some tests are disabled and forcing outcomes does not work

Dunno why it was ever `complex`

Added some more terms to clifford.by_name to ease lookup of complex
conjugates

Added GraphState.toggle_edges(), used by GraphState.measure_x

Nearly finished GraphState.measure_x