Serialize systemd jobs properly

[sukrit007]

Fleet currently does not apply locking when scheduling conflicting operations like "destroy", "start". For instance:

fleetctl destroy
fleetctl start

It might happen, that the fleet might be in process of stopping the container, while start might run concurrently.

If it runs on separate container , it results in issue :
#638

When it get scheduled on same machine, it results in docker issues ( as you might be in process of stopping and starting a docker container at same time which docker does not support).

Overall , I think we need some kind of locking abstraction in Fleet, so that fleet should not run the start command if any other conflicting command like stop is in progress.

[bcwaldon]

To add a little bit more flavor to this: fleet is always submitting systemd jobs with "replace", which means any running jobs are shot. A StopUnit, which is in the process of gracefully shutting down a docker container, could be stopped prematurely by a StartUnit and put docker in a bad place.

[jonboulle]

related: #669

[bcwaldon]

Fixed by #755