Fixes #54

Obviously this needs an integration test, but I wanted to just get implementation structure now. It is functional, tested with the demo script.

ERROR:dispatcher.worker.task:Worker 1 task canceled (uuid=1db13c11-a50a-4b0a-91c7-8cbc401c7436)
Traceback (most recent call last):
  File "/home/arominge/repos/dispatcher/dispatcher/worker/task.py", line 84, in run_callable
    return _call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^
  File "/home/arominge/repos/dispatcher/tools/test_methods.py", line 28, in task_has_timeout
    time.sleep(5)
  File "/home/arominge/repos/dispatcher/dispatcher/worker/task.py", line 31, in task_cancel
    raise DispatcherCancel
dispatcher.worker.task.DispatcherCancel
DEBUG:dispatcher.pool:Worker 1 finished task (uuid=1db13c11-a50a-4b0a-91c7-8cbc401c7436), ct=16, expected cancel, canceled

There are some aesthetic changes we could make, like call it "timeout" instead of "cancel"... but obviously timeout uses the same mechanism as cancel. Maybe I'll just rename the exception to DispatcherInterrupt or something like that.

Also, I'm relatively happy with the structure, although it took some thought. I implemented "delayed" task in the "main" part of the code. It's non-obvious that timeouts will go in the pool part, but the pool can gate on capacity, and timeouts don't start until the worker actually starts running it. And this is definitely something we want enforced in the parent, in general.