Worker: looking a bit fragile if execute throws

The worker's top level execute function has a few places where it might potentially throw.

This would result in the worker crashing and all runs in progess getting lost.

This is obviously quite severe but it's not entirely clear what we should do. The worker has an obligation to try and finish all outstanding work and report back to lightnig.

For example: failure to join the run channel will result in an unhandled exception. At this point something is very wrong. I don't think we can report back to Lightning yet because we've not even been able to join the comms channel (or we were kicked out). So the run is probably lost at this point. We should probably kill the whole worker (but give existing runs a chance to finish)

There are probably a few other equally nefarious places where an error will result in the whole worker crashing. It's a very sensitive bit of code with a lot of asyncrhonicity and dependencies on other systems.

Start at the worker's top-level execute function and trace every operation that can throw, including joining the run channel and reporting back to Lightning. Determine the required behavior for each failure, especially how outstanding runs should finish or be handled when the worker loses its communications channel; done requires an agreed failure policy and coverage of the identified paths.