node_features/knl_cray plugin: drain any node not reported by
    "capmc node_status" on startup or reconfig. Also re-tests
    on failed node restart for job.

node_features/knl_cray plugin: Remove any KNL MCDRAM or NUMA features from
    node's configuration if capmc does NOT report the node as being KNL.
    For example, we don't want a non-KNL node with features="quad,cache".

Found by clang.
Continuation of 76d62ae4

Add new knl.conf configuration parameter CapmcRetries
Modify capmc_suspend and capmc_resume to retry operations when
  Cray State Manager is down.
Add retry logic to node_features/knl_cray to handle Cray State
  manager being down.
bug 3100

from commit ee4a9776.

reset by list_count.  Also remove a nested if for cleaner code.

Logs go to both locations when running in non-daemonized mode.
Don't refer to this as "debug" mode, while useful for debugging
it's not directly related.

Bug 3146.

Otherwise they'll truncate when packed into the RPC and end up
as some bizarre value at the controller.

Bug 3098.

Set completed time for pending/running runaway jobs to the max of
(start, eligible, submit) times.

Bug 3075

Avoid using slurm_forward_data because it causes thread spawn that
introduces unwanted delays.

Bug 3102.

Fix indent and add brackets

Also correct the value of NICE_OFFSET used within the perl API.

Bug 3098.

Bug 3051.

Otherwise updates would be rejected for running jobs even if there
would be no impact. Most common when the job_submit plugin is overriding
QOS/GRES values on everything; without this change an update to "comment"
or other fields would fail with ESLURM_JOB_NOT_PENDING.

Bug 3117.

Never ever run NHC, even on an edge case that NHC_NO would still launch NHC
after.

Bug 3105.

Switch to list_for_each, and check if access list actually changed
after each update before updating last_prat_update. This prevents
the backfill scheduler from resetting mid-cycle unnecessarily.

Bug 3123.

Buffer was being allocated based upon the size of the wrong structure.
  The buffer would have been larger than required, so this bug
  should not have caused any failures.
bug 3127

Added memory limits to the job and step. Without these, only one
  step may be able to run at a time and break the test

Add "sbatch_wait_nodes" to SchedulerParameters to control default sbatch
    behaviour with respect to waiting for all allocated nodes to be ready for
    use. Job can override the configuration option using the --wait-all-nodes=#
    option.
bug 3120

Prior logic would treat execute line like this:
$ sbatch --wait-all-nodes -N3 tmp
with "-N3" as being the argument to the "--wait-all-nodes" option.
See bug 3120

Add salloc/sbatch/srun option --use-min-nodes to prefer smaller node counts
    when a range of node counts is specified (e.g. "-N 2-4").
bug 2996

Switch a few SLURM mentions for Slurm as well.

It was out of alphabetic order before (e.g. after --power).

Add salloc/sbatch/srun --priority option of "TOP" to set job priority to
    the highest possible value. This option is only available to Slurm operators
    and administrators.
bug 3115

The problem reported is just a configuration warning and not an
 error. Also change the test from ">=" to ">".
bug 3086

This patch finally resolves absolute/relative CPU mapping for nodes
where the NUMA (or sockets) have different core counts (e.g. KNL SNC4).