Cleanup in case of timeout or job cancelation

[morcuended]

The current approach does not work in error situations such as job cancelation (e.g., due to timeouts). We want to be able re-launch the processing without having to remove the created but spoiled files manually.

It is possible to handle the system signals raised by SLURM when a job is canceled and use them to clean up afterward (see e.g., https://www.hpc.caltech.edu/documentation/faq/how-can-i-handle-system-signals-my-job)