Node failure job with no `return_code` file still marked as successful in Galaxy

Running hifiasm on the VGP instance and the node it executed on appears to have failed and the job was terminated prematurely. As a result, nothing past a certain point in the job script was executed, including capturing/writing the tool return code to `return_code`. Despite this, none of the postprocessing failures were fatal (due to being work_dir outputs) and for some reason, Galaxy still treated this job as successful.

Trace how Pulsar handles a terminated job when the `return_code` file is absent, including whether failures in `work_dir` outputs affect the final status. Reproduce the hifiasm scenario if possible and verify that a job whose execution stops before writing `return_code` is reported as failed rather than successful.

Written by the indexing model from the issue text.