Improve the actions and their test coverage

[JulianTrommer]

Currently the build and drive action only check if the image can be built and what the current driving score is.
It is not covered if a node of the agent crashes or something went wrong with the agent.

There should be another action that checks if all nodes are started correctly.
Another consideration is if the drive action should be called manually instead of with each pull request. That way PRs that do not improve the agent don't block the runner pipeline.