Experimental/what if gitlab fails in early stage

• Test to see if the on failure stage will help

• Make a file to track if we failed or not

• Allow jobs to be interrupted

• Add magic file method to other jobs

• Fix standard test case broken for testing failure state

• Make better use of the special file used to report the result

Whilst we’re touching the gitlab yaml anyway try to improve the handling of failure at various points. Now essentially use JOB_STATUS_FILE to contain our current view of how the job is going (building, failed, passed etc.) and then use the after_script stage to push the final state of the job. I think this should ensure that jobs are always updated on Bitbucket whatever stage they fail at, which is nice.

Also fixes OpenMPI + docker related issue that was causing a large amount of noise in the logs ( see https://github.com/open-mpi/ompi/issues/4948 for details).