Non-deterministic job execution

[henricasanova]

Because when creating a standard job the pre-file copies, post-file copies, and post file deletions are in sets of items, and the order depends on the memory allocation. So depending on the allocation (and perhaps from run to run), a job execution can differ, which is not great.

The idea would be to modify the API and replace these sets by vectors, so that there is a consistent order.