heptapod-paas-runner: better error treatment when provisioning fails
I had a connection error on the Clever Cloud API earlier today:
Dec 22 18:58:44 heptapod-foss heptapod-paas-runner: requests.exceptions.ConnectionError: ('Connection aborted.', RemoteDisconnected('Remote end closed connection without response'))
In itself it would warrant a retry, but there is a more severe problem:
Some, maybe all, later jobs got acquired from the coordinator, but no provisioning attempt occurred for it (and no error). It really looks as if the thread responsible for that died. A service restart did fix the problem, but those jobs were lost.