This should no longer be needed on the slurm cluster, where each task
can request the amount of tmp space it requires explicitly.

Removing the tmp from the shared filesystem back onto the host running
the analysis should also improve the performance of the pipeline.