Document multi-node with Dask on Databricks

[jacobtomlinson]

Our current Databricks documentation shows how to launch a Databricks cluster (single and multi-node) with the RAPIDS environment. However it doesn't actually discuss how to leverage the multi-node cluster with Dask.

We need to document this.

I expect this will involve using dask-yarn to start the scheduler and worker processes and will need to use dask-cuda-worker as their startup script. Databricks doesn't use Yarn, we would need to explore whatever it does use.

Related issue #231 covers documenting the single-node setup.

[beckernick]

Should we file a separate issue for documenting how to use a single-node Dask cluster on Databricks?

Should it be as simple as spinning up a LocalCUDACluster on the node? A quick Google search provides few results/examples.