Add retries for analysis jobs

I've noticed an increasing number of jobs failing at CIT due to file system issues, mostly due to "no space left on disk".

I think that one workaround for this is to specify some number of automatic retries.

@gregory.ashton are you in favour of having this?