heptapod-paas-runner: keep track of decommissioning to perform
We currently don't have a set of decommissioning to be done. As a consequence, if there is a shutdown (or restart) after a job is identified as finished yet before it has been decommissioned, the resource will linger, and undue billing will occur.
See also #28 (closed), for which we'll add a retry (and sleep) logic to decommission.