Resolve "Add GPU support for slurm environment to bedpostx"

Closes #2

See merge request !2

This is hard coded for now but allows bedpostx to specific the
correct parameters for the fsl_sub job submits for gpu use in
slurm.

Allows caller to pass additional information to scheduler.

This lets the caller of fsl_sub enrich the job submit with context
specific arguments, eg. request a GPU.

This influence the slurm partition used with -T (time) arguement to fsl_sub.
Enables caller to select correct queue.

Need to explictly set a SLURM detected var since there
is no inherent var like SGE_ROOT for SLURM

This supports correct mailopts for slurm and avoids errors in the
fsl_sub command.
Tag the changes with SLURM as well as SGE

Bring over changes from slurm support developed in bedpostx.
These are minor corrections to variable values and conditionals to support
SLURM along with SGE.
Additionally include the cancel script customization for slurm.

Merge branch '2-add-gpu-support-for-slurm-environment-to-bedpostx' of gitlab.rc.uab.edu:jpr/fsl into 2-add-gpu-support-for-slurm-environment-to-bedpostx

Allow the -q parameter to define the partition for slurm.
Needed for GPU use.
Set default queue value to the previous hardcoded value "short".

Reuses FSLGECUDA environment var for GPU partition definition in
slurm.

This is the default wrapper for GPU execution provided by FSL
and designed to work with SGE.

Detect slurm or sge and use appropriate mail sending options.

Provide the same support for job control that SGE has
and exit with job control report.

The -m parameter is not always set by the caller so set the proper value
from ldap by default.

The slice jobs don't need as much ram because the image
slices are smaller.  The factor is about 10x the file size.
1GB is a generous amount.

The "afterany" is the correct dependancy to hold jobs until the
prior job completes.

Execute task in sub job not in shell of caller.

The slices run for about 25min/megabyte in the slice file so 12hrs
is a safe max.

Slurm log file pattern added for array job to move output to log dir

This completes the prior commit.

Override the hard-coded FSL upstream email address structure and
replace with a local ldap query to get working email for job
notifications.

Added SLURM array job submission modeled after  the SGE array job structure
Removed the stdin processing for command list which doesn't work
Slurm returns job id for caller to use to create dependency tree
Added variables similar to the sge_ vars for slurm_ to get the arguments
needed for job submission.

Applied from fix upstream patch http://www.tchpc.tcd.ie/node/439

patch fsl_sub fsl_sub.patch

This is the patch from http://www.tchpc.tcd.ie/node/439
which provides an outline for adding slurm support to fsl_sub.
Should run with `patch fsl_sub fsl_sub.patch` but throws an error.