Currently, running pipe_drivers tasks using SLURM gives one log file saved directly by lsst.log as job_$JOBNAME.log and another log file saved by the batch system catching the stdout/stderr. This seems unnecessarily confusing.

IIRC I first added it to ease the transition from using pex_logging only, to mixing pex_logging+log, and to using only log in the pipelines. Now pex_logging is no longer used, maybe we only need to keep one of the two?

I like Paul Price's idea at DM-8359:

Maybe we should just remove the log file altogether, and rely on the batch system catching the stdout? Then we need to make sure that the host and PID are being prepended to the log messages.