launch reply to srun.

possibility of killing the wrong job if killed jobs somehow avoid getting
their record updated in shared memory.
(gnats:300)

   verifiers. (Allows a single re-use of given job credential)

   entry.

    service_connection()
 o Check to see if fd already closed (fd < 0) before closing again
    in service_connection()

  - install atfork handlers in slurmd to protect against fork-critical
    code (gethostbyaddr and list_*)
  - do not bother freeing slurm credential context
 o Fix small bug in save_cred_state
 o Log more errors in slurmd manger and session mangers
 o change setproctitle and logfile prefix for NOALLOC jobs

equivalents. For consistency make them return integer values (like before)
that are not valid characters (e.g. 0x100). This avoid the use of single
character values that we might want to preserve.

Job's priority would be reset if it could not be immediately scheduled
AND if the best-fit node selection picked a DOWN node.

client and server are both reusing ports that match a defunct socket
connection in TIME_WAIT state. On this failure, we select a new random
port number and try again to connect.
(gnats:253)