No visible error when job runs out of memory

[donkirkby]

The fix for #758 marks runs as failed after a job gets killed, but there's no error message indicating why the job was killed. Particularly when a job is killed for using too much memory, that would be helpful for the user to see.

ContainerRun.check_slurm_state() should load the Slurm stdout and stderr files if a job dies.

This will probably be closely related to the fix for #785.