Some fixes for SLURM and SGE engine

[michelwi]

While working on the engine for a different project, my LLM pointed out some problems:

• SLURMs job_id is a list[str] instead of a single str as it is for all other engines
• Off-by-one error in the range that specifies which task ids have been submitted. In sbatch end_id is inclusive, for range it is exclusive.
• The check if a jobs rqmt has been updated is broken since we add a bunch of other stuff to the submit_history so rqmt_hist[0] == rqmt is never true. Instead, I now compare if rqmt_hist[0] is a subset of current rqmt and if not only update the changed keys.

This changes the content of submit_log.<task>.<id> but since we don't use this file for anything anyway, I decided against including compatibillity code.

before

([1], {'cpu': 1, 'mem': 10.0, 'gpu': 0, 'time': 1.0, 'sbatch_args': [], 'engine_info': [([], ['26316813'])], 'engine_name': 'slurm', 'completed_fraction': None})

now

([1], {'cpu': 1, 'mem': 10.0, 'gpu': 0, 'time': 1.0, 'sbatch_args': [], 'engine_info': [([1], '26316813')], 'engine_name': 'slurm', 'completed_fraction': None})

(note the changed engine_info)