Currently stdout for a job is stored in the job_instance table. This has been a problem in the PMC case because the stdout is so large.

It would be better if we stored stdout/stderr in the invocation table. That way we can split up the stdout/stderr for clustered jobs and make it more useful for users.

We would still need to support stdout/stderr at the job_instance level for the cases where Kickstart is not used.

It may be sufficient to simply store references to the stdout/stderr files at the job_instance level.