Slurm, fix duplicate jobs

[albertz]

Fix #156

[albertz]

Can someone review?

[michelwi]

The behaviour looks correctly implemented to me.

Discussion question: should we add a flag to enable it? should it be on by default?

Also @albertz did you test this with the i6 cluster and does it solve your problem?

[albertz]

Yes I tested this, and I did not see the problem anymore (but maybe the cluster was just behaving fine since then).

In any case, I think if slurm responds with a valid job id, and also has return code 0, I don't see why we should ignore and throw away that job id. This looks like a bug to me.