heptapod-paas-runner: persistent state for decommissions, avoiding full block

We move first decommissions out of the event processing thread, preventing it to crash.

Then we introduce state persistence for jobs to decommission.

Closes #28 (closed)