Support taking node out of FUTURE state with "scontrol reconfig" command.
Previous logic would keep node in FUTURE state if that was the original
configuration when slurmctld started. If job was running on the node,
it will stay running, but the node make not be visible.

Make SLURM_ARRAY_TASK_MIN, SLURM_ARRAY_TASK_MAX, and SLURM_ARRAY_TASK_STEP
    environment variables available to PrologSlurmctld and EpilogSlurmctld.

We needed to send a finish from each node in the step whether it had
any activity or not.  This way the controller knew things were done
on the node and the data was sent to the database.

Bug 2097

to get the batch step, no real code change outside of using strcasecmp
instead of strcmp.

Burst_buffer/cray: Don't stall scheduling of other jobs while a stage-in
    is in progress.
bug 2114

Fix to purge terminated jobs with burst buffer errors.
bug 2123

The prolog_running counter can now exceed 1. New logic raises limit
from 1 to 4 before preventing job recovery on restart.

the error happened.

Correct preservation of job ID. This effects emulation mode only.
bug 2113

Added burst_buffer.conf flag parameter of "TeardownFailure" which will
    teardown and remove a burst buffer after failed stage-in or stage-out.
    By default, the buffer will be preserved for analysis and manual teardown.
bug 2116

If a stage-out fails, Slurm leaves the burst buffer in place and
  logs something like
  "error: bb_set_use_time: job 98 with allocated burst buffers not found"
  every minute thereafer.
  This changes the logic to only log the event one time.
bug 2112

This is a revision to commit e21b666c
which did not fix the problem for all configurations.
bug 2086

Bug 2106

What was happening was the calculation wasn't happening for memory
or nodes, just cpus and gres.

bug 2086

Job needs exit code of zero for COMPLETED state.

bug 2093

This reverts commit 5bf40e44.

pre-increment.

The "dw_wlm_cli paths" command returns an error if no #DW options.

commit:508f866e

Conflicts:
	etc/slurmd.service.in

Bug 2095

Increase dead socket timeout from 5 to 10 minutes before purge
Log connect failure as "debug" rather than "error" as it can be
  generated due to a race condition at step termination.

Previous logic would not check on remaining persistent burst buffer
space needed for a job.