This solves issue #2338 with a simple two-line fix. I tested it with dry_run_level=1 on two separate nights, which were both successful. Those tests should provide full coverage for the generation of a processing table row. The code no longer crashes when requesting pernight redshifts, and the pernight job rows have unique internal ID's and the correct metadata.

This PR also removes the desi_run_night but not the underlying Python code, just in case we want to bring it back in the future (though that is unlikely). This will prevent people from using it accidentally instead of the new and similarly named desi_proc_night.