Overlapping execution of run sets

[PhilippWendler]

If a benchmark definition has several run definitions, they are executed strictly sequentially. With parallel execution of runs we could let them overlap, such that the first runs of the second run set are already executed while the last runs of the first run set are still executing but some resources are free already.

The main difficulty to do this is that our OutputHandler (which writes the results to XML and stdout) currently needs to have one run set finished before the next one can start. This class would have to be refactored. And we would need to think how the stdout output should look like such that the user is not confused.