Add ability to run "one shot" commands

[soenkeliebau]

The idea here is to support running preparatory or maintenance tasks that need to execute as a command that finishes after a while.
Ideally this should also be done via systemd as some of these commands can run for quite a while (days sometimes) and we do not want them aborted, when the agent crashes or is restarted.

This could be done via the following settings:

Type=oneshot
RemainAfterExit=true

An example can be seen here.

I am unsure if RemainAfterExit is necessary, but this might be required to check the exit code of the command afterwards.

[siegfriedweber]

Kubernetes jobs behave as follows:

• The job controller creates a pod with the label job-name=<name of the job>.
• If the pod terminates successfully then its status contains phase: Succeeded and status/containerStatuses/state/terminated contains exitCode: 0 and reason: Completed.
• If the pod terminates unsuccessfully then its status contains phase: Failed and status/containerStatuses/state/terminated contains exitCode: <exit code> and reason: Error.
• The job does not delete the pod so that the result can be inspected.

(The job controller actually does a lot more but that is not relevant here.)

This behavior should be replicated in the agent.

[lfrancke]

We have this ticket stackabletech/documentation#28 which talks about designing this feature.
That has been superseded by this one.
Is this a thing/change that needs to be documented? I believe so, right? We should document the Agent behavior.
I'll close the other issue in favor of this one.