slurmctld sends REQUEST_KILL_TIMELIMIT when job reaches its time limit,
change job state to TIMEOUT | COMPLETING and node state to COMPLETING.
slurmd upon receipt of REQUEST_KILL_TIMELIMIT sends job SIG_XCPU and SIGKILL
then runs epilog.

slurmctld sends REQUEST_KILL_TIMELIMIT when job reaches its time limit,
change job state to TIMEOUT | COMPLETING and node state to COMPLETING.

Add COMPLETING node and job states.
Major restructuring of slurmctld node and job state transition code.
Change of comment in srun.c

Add COMPLETING node and job states.
Major restructuring of slurmctld node and job state transition code.

Add COMPLETING node and job states.

 o rename modules to not have "lib" prefix and install them into pkglibdir

of "fprintf(stderr,...)").

 o src/slurmd : if session manager process is killed, ensure that entire
   session is also dead, and make sure all task exit msgs are sent to
   clients
 o src/slurmd/reattach.c : initialize job->hostid in reattach code

   of running tasks in error message for task exit.

 o srun now prints exit status more like the shell, and groups output
   on a single line as much as possible.

 o remove code from no_interconnect.c that is now used by default in
   src/slurmd/smgr.c : _setup_env()

 o src/srun : move report_{job/task}_status into job.c
              report task exit on one line per msg (instead of per task)
              add new task state for abnormal (nonzero) exit.

   (possibly significantly) on network traffic.

   that job/job step session leader runs as owner of job
   and not root.

   Split "session manger" code from mgr.c (which runs as root)
   into smgr.c (most session manager code runs as user)

 o reorganize interconnect functions (see interconnect.h) to
   fit new dual-process model for slurmd.

 o add pipe to job structure for communication between slurmd
   job manager and session manager.

 o add "mpid" field to shared memory for job steps and
   shm_update_step_mpid() to update this information.

 o add ulimits.[ch] to support  setting user limits for jobs based
   off SLURM_RLIMIT* env vars.