Throw an error if enabled but the build had it disabled.

Reject jobs with X11 flags set if the support is not enabled,
and add an appropriate error code for this case.

- Do not re-launch the extern step if !first_run.
- Move before _remove_job_running_prolog() to make sure the
  extern step has at least a slight head start before steps can
  launch. Note that this only ensures the stepd has started, but
  not that the stepd is finished it's own internal setup - further
  work in slurmstepd would be required to remove this race completely.

Bug 4264.

Return an error if something doesn't work out.

Need to split this back out to have a chance to encapsulate the
extern step launch in _rpc_prolog as well.

This is a repeat of a block later on, at a point when
rc could potentialy be something other than SLURM_SUCCESS.

Ensures that slurmctld can centrally manage availability of the flag.

Of course, x11 forwarding needs to be setup as part of the prolog,
but this will avoid expensive calls to the extern stepd to lookup the
display if the job definitely does not have x11 forwarding enabled.

This commit is here to get the Makefile changes in place ahead
of the implementation. Only build these if the native X11
forwarding is enabled (which implies libssh2 is functional as well).

Coverity CID 44985

Coverity CID 44974

Coverity CID 44801

Coverity CID 44781

Bug 4138

Avoid using configured SallocDefaultCommand and failing test

Required for test to make sure the time limit is not adjusted by
  the backfill scheduler

Bug 4276

A pack step may have multiple steps connected to STDIN/OUT/ERR for
  I/O. When a task ended, it would close the file, resulting in an
  I/O error for the other job step components. This change avoids
  closing STDIN/OUT/ERR.

Expand some log messages
Rename a local variable from "info" to "read_info" to avoid conflict with
   our logging function name
Change some comment layouts to conform with Linux kernel standard

Bug 4244

to handle case insensitive ClusterNames and handle spaces before equals.

Take the best concepts from both functions and produce a unified
version so now the list of accepted signal names will stay in sync.

Also fix the comment to note that 0 is returned on failure, not -1.

Bug 4227.

Also modify a log message for better clarity

Previous logic would result in error if Slurm version > 17.11

The problem was the tests set the same variable, opt_arg, for for
options, --test_suite and --test_suite_reg. But with the addition of
de24ce72 the variable was being set with the different numbers out
of order because the variable was being set from the environment. So the
fix is to have the options set their own variable.

Bug 4176

to avoid name collisions with c++ list types. gdb started showing the
wrong type when the full lib was created.

to set -O0 since --enable-developer and --enable-debug set
optimizations.