Add a SLURMRunner

[jacobtomlinson]

This runner could use the environment variables set in each process for self identification and a shared filesystem for communication.

Each process can use the SLURM_NODEID for it's ID (a unique monotonic index that is assigned to each process) and the SLURM_JOB_NUM_NODES to know the total number of processes.

• The process with rank 0 assumes it is the scheduler, it writes a scheduler file to the shared filesystem.
• The process with rank 1 assumes it should run the client code, it waits for the scheduler file to exist and then continues running the contents of the context manager.
• All processes with rank 2 and above assume they are workers, they wait for the scheduler file to exist and then start worker processes that connect to the scheduler.

[lgarrison]

I'd be interested in this functionality, could I help out? Is the project ready for contributions? I'm no Dask expert but I think I understand the relevant pieces here.

(As a minor point, I think the relevant Slurm variables to index processes are actually SLURM_PROCID and SLURM_NTASKS. Ref.)

[jacobtomlinson]

Closed in #3