Investigate more autobahn failures and improvement to the test

[mstoykov]

There are still some test (from the very late ones) that are failing. Unfortunately it isn't immediately obvious what they test or what is failing exactly so that will likely take quite a lot of work on it's own :(.

It might also be due to the websocket library we are using which is still looking for maintainers and isn't all that up to date.

We likely will want to at least try other variants in the future as well.