Better canceling of job at the beginning of the run.

[arturopie]

Jobs are not cancelled until all workers start running files.

Despite the originator finishes right away, listeners, worker managers and workers keep running (calling callbacks) until they poll from file queue. As a result, cancelling a job at this early stage will not properly clean all queues. Besides, the remote processes will do work that is not needed (callbacks) and since the Originator finishes right away, the user will have the impression that all the machines are ready to run another job.

Rewrite 'cancelling a job', so that worker managers can check if the job was cancelled before they run before_creating_workers callback, and before forking workers.

It would be nice if the listener can also receive the cancel_job message, so it returns before creating worker manager

I would also be nice if the originator waits for all worker managers to finish before originator finishes.