I thought about this issue and I would say that it's better to keep building and 
starting of an experiment separate. The experiment.sh wrapper simplifies 
submitting the experiments and one only has to change into the experiment's 
directory if one wants to submit the experiment manually. Therefore I am setting 
this issue to resolved.

Even if there is an option to directly start experiments, I think we'd still
need a way to handle experiments that are not immediately started in a better way.

I thought about this a bit yesterday, and I think we could/should add some
path-guessing logic into the code. (That would be a separate issue, though.)

This eliminates the need for or is supposed to be a qsub wrapper.