Pipeline_base add functionality to rerun downstream if upstream is run.

[HenrikSpiegel]

add a flag so that the pipeline will delete success_files of downstream jobs if an upstream is rerun.

This should avoid having "desyncs"