Made sure that `graph.js` dumps old graphs after use, and ensured that
`examples/stress_test` shuts down the connection after use (don't create millions
of clients).

By default the server echoes messages to all clients.
The fancy.GraphState client was never calling ws.recv()
And hence a buffer was filling up somewhere
And we were hanging.
Adding a ws.recv() fixed the problem.

Got rid of all the debug harness and refactored graphstate.py

Next up: measurement
Also added a test on local complementation