Fix DRMAA FAILED job state reported as COMPLETE

[gkr0110]

Summary

BaseDrmaaManager._get_status_external() maps JobState.FAILED to
status.COMPLETE, with a comment already acknowledging this is wrong:

JobState.FAILED: status.COMPLETE,  # Should be a FAILED state here as well

Any job that fails at the DRM submission or execution level (e.g. rejected
by Slurm for a missing/invalid account, or any other DRM-side failure) is
reported back to the Galaxy client as a successful completion, with no
output and no indication anything went wrong. We hit this concretely: jobs
submitted via queued_drmaa to Slurm through slurm-drmaa that got
CANCELLED for lacking --account were reported to Galaxy as complete
with an empty output directory.

status.FAILED already exists and status.is_job_done() treats FAILED
and COMPLETE identically for job-done detection, so this change only
affects the terminal status surfaced to the client, not job-done detection.

Test plan

• Confirmed via git blame this mapping (and stale comment) has been
  unchanged since it was introduced.
• Grepped test/ for references to _get_status_external /
  JobState.FAILED -- none found, so no existing tests assume the old
  (incorrect) behavior.
• Reproduced against a real Slurm cluster via slurm-drmaa: a job
  submitted without an account is CANCELLED by Slurm and previously
  surfaced to the client as complete; with this change it correctly
  surfaces as failed.

[jmchilton]

Amazing - thank you for such a complete description of the issue, testing, etc.. and thanks for the fix! My agents found some polish they wanted to do at the higher layers and other managers based on this fix. Merging that PR should merge this one too - thanks again so much - great work! -John

[gkr0110]

@jmchilton - It’s a tiny tweak, but I think it could be helpful for others. I spent a couple of hours trying to figure out what was breaking..

We are building a new Galaxy instance for Switzerland, so testing out the Pulsar workloads heavily. So expect more PRs/Issues 😄