This eliminates jobs as soon as possible from the queue.
Also a bad job state reason in the new logic is fixed.

If the slurmstepd message to slurmd notifying it of batch job completion
occurs before the slurmd processes it's launch response, then the slurmd
would revoke the job credential, resulting in the launch response generating
a second message to slurmctld. This patch eliminates that second message
for select/serial.

Build the batch_job_launch_msg_t in one function so it can be used
to launch a new job via agent.c and another for use as a select/serial
batch job completion response

This eliminates one of four RPCs needed for each job.

Do not wait for slurmctld to request job termination

This will eventually permit the slurmctld to respond to the slurmd
with a new task launch RPC