Get rid of job_name_set_cmd/job_name_set_env in srun_opt_t and use
slurm_option_set_by_cli()/slurm_option_set_by_env() instead.

Add this as an option to srun; there's no reason it can't use
this as an alternative to --nodelist.

Move the file parsing code to later in _opt_verify(). This should
fix a potential source of problems if an #SBATCH directive points
to a non-existent file which is being intentionally overridden on
the command line.

Note that this intentionally discards the checks for a nice
value < 0, which were previously limited to SlurmUser / root.

However - an Admin is allowed by slurmctld to submit with a lower
value, or they can alter the job after submission to change the nice
value further. So get rid of that check here.

Use -1 as the sentinel value, adjust srun to match salloc/sbatch.

Note one difference after the change: an invalid optional argument
to the flag will be rejected, rather than silently turning on --no-kill.

Change cwd variable in slurm_opt_t to chdir to match argument.

One minor change - an invalid environment variable will result in
an exit(-1), rather than an error message.

Reset on each pass for sbatch, but not touched in salloc/srun.
Go with sbatch's behavior, which matches that of mail_user.

Change field name to power in slurm_opt_t, and update all locations.

Remove casts through (char *) now that power_flags_id() takes a
(const char *) as input.

The salloc command did not reset on each pass - or ever - so follow
the established behavior of sbatch/srun and reset for each component.

Note that invalid env vars will result in exit(-1); instead
of just an error message.

Drop it from sbatch since it just printed an error if you tried to use it;
getopt_long() will happily tell you that option isn't available itself.

Change slurm_opt_t field to 'exclude' to match command line.

Node that the _valid_node_list() call is being moved into _opt_verify()
where it probably should have been before. Since _valid_node_list()
depends on several other settings, processing it when --exclude is
processed may have resulted in inconsistent behavior if the order of the
various arguments was changed around.

Note that srun reset this only once, while salloc/sbatch reset
each pass. Go with salloc/sbatch behavior - I do not think
carrying the distribution over between components makes sense
for most allocations at this time.

Note this overlaps with the --exclusive option, but inverts
the default behavior if no arguments are given.

Note that srun was previously using NO_VAL as unset which is technically
an acceptable value; change it to -1 to match salloc/sbatch.