Make complexity graph history more robust

Currently, the history in the complexity graphs is implemented via downloading the artifacts from the last job of the same type. This is not very robust since the artifacts are lost when a job fails. Ideally, the complexity data should be stored in some sort of central persistent storage.

Possible solution: Set up a dedicated branch/repo and have the compelxity jobs push their data there. Checking out this branch then will automatically give the latest data. This could also be used to separate the data generation from the presentation.