Remote Jupyter kernels on Dask cluster

I've had a thought that I'd like some feedback on before I investigate further.

Would it be possible to start a notebook kernel on a Dask worker and connect to it from Jupyter Lab? Using an approach like [this](https://github.com/ipython/ipython/wiki/Cookbook:-Connecting-to-a-remote-kernel-via-ssh), but not necessarily using SSH to proxy the connection.

Ensuring environments are consistent between client and workers can often be frustrating, especially when a user has Jupyter Lab in a local conda environment on their laptop but is launching a cluster somewhere like Kubernetes where the workers will be using a Docker image.

Running a kernel on a worker may be one workaround/solution to this.

I would imagine the workflow to be something like:

- Open Jupyter Lab
- Launch Dask cluster using Lab Extension
- Extension adds the cluster to the list of available kernels
- User can launch the kernel:
- From the launcher
- By opening a notebook and changing the kernel to the Dask cluster
- Directly from the cluster in the Dask Lab Extension
- Shutting down the cluster would remove the kernel from the list

@ian-r-rose @mrocklin