Previous logic would deadlock on fork if log_init() was followed by 
log_fini() since log_fini destroys the mutex that atfork works with.

 needed thanks to slurmd_step functionality.

still open.  Each stdout or stderr that closes will trigger an
eof message (header.length will be 0), and _server_read will
decrement the number of remote writers.

is a pointer.

This merge is a rewrite of slurm's task standard I/O handling.
srun will now have just a single TCP stream to each node of a job-step.
srun and slurmd comminicate over the TCP stream using a simple messaging
protocol.

Both srun and slurmd now use a completely event driven I/O engine based on
the eio.c event I/O engine.

Be warned that the new I/O support in this merge is not quite complete.  The
core I/O protocol is functional, but it is still missing reattach support
and it needs extensive testing and debugging.

   TaskProlog, TaskEpilog, TaskPlugin; srun --task-prolog, --task-epilog).

    REQUEST_KILL_JOB/TASKS changed to REQUEST_SIGNAL_JOB/TASKS

    slurmctld (at least for linear selection algorithm).