slaunch will now cancel a job step on SIGINT, SIGHUP, and SIGTERM.  If
a signal it caught, it will signal the step to terminate and wait up
to 10 seconds for all remote nodes to report that the tasks have exited.

After 10 seconds, it will give up waiting and exit, regardless of the
state of the job step.

step_launch api code.

It now allows a comment on the line, quotes to group characters together
(to allow whitespace in arguments), and a backslash escape character
to allow inclusion of #, \, ", etc.

slaunch) from src/api/step_pmi.[ch] to src/api/pmi_server.[ch].

Also, delete srun's personal copy of the pmi server code and have it
build against the copy in the shared src/api location.

Overhaul the command line parsing code to handle #SBATCH arguments in
  batch scripts.

Remove unnecessary multi-prog option (multi-prog is slaunch-only).

code from test18.14.

with and associated launcher.  A batch script only has job allocation-specific
variables set.

Add new SLURM_JOB_* environment variables.

variables are missing.

in the spec file did not have any parameters.

Also, no need to sift through environment variables for the task's ID,
the slurmstepd already knows that.

There appears to be a bug in slurmstepd's task_exec.c parsing of the
multi-prog script when commands have no command line parameters.

srun is also broken in this respect, it is not slaunch specific.

honoring the overcommit flag.

The "--threads" option is now named "--pmi-threads", since it only applies
to PMI threading.  At this time slaunch has no other threading which might
requiring throttling.

made step_spec->overcommit 0 out cpu_count when creating a step so we don't check for avaliable processors. making it overcommit