No queue arguments needed on SLURM for most clusters.

[bw4sz]

Removed queue from default arguments. The default was empty, which caused the following arguments to be poorly parsed by SLURM. The university clusters I have worked with do not use the -p flag. Of course, it may vary by cluster. This reverts #18, which probably wasn't needed.

Closes #15

[bw4sz]

Confirmed works on University of Florida HPC environment.

(pangeo) [b.weinstein@gator3 dask-jobqueue]$ python test_slurm.py 
(pangeo) [b.weinstein@gator3 dask-jobqueue]$ squeue -u b.weinstein
             JOBID PARTITION     NAME     USER ST       TIME  NODES NODELIST(REASON)
          18337278 hpg2-comp     dask b.weinst  R       0:14      2 c31a-s[4,21]
          18337279 hpg2-comp     dask b.weinst  R       0:14      1 c22b-s42
          18337280 hpg2-comp     dask b.weinst  R       0:14      2 c22a-s36,c22b-s40
(pangeo) [b.weinstein@gator3 dask-jobqueue]$ cat test_slurm.py 

from dask_jobqueue import SLURMCluster

cluster = SLURMCluster(project='ewhite')
cluster.start_workers(3)

[jhamman]

Let's put this on hold until after #10 gets merged. I think there will need to be some minor modifications to the SLURMCluster but they should come after #10.

[guillaumeeb]

Yup, for consistency, we will have to refactor SLURMCluster as mentionned in #22. As for this pull request, I'm not entirely sure removing queue keyword is a good thing. Don't know Slurm, but I believe being able to specify a queue is a good thing.

After #10 is merged, we may put the queue keyword as None by default, and by building the job_script dynamically, we could avoid add the -p option being positioned in the script.