Enhancements about custom kernels and relevant documentation

[whyjz]

As per the discussion thread on Slack, here are some ideas we could consider to improve the cloud-based research activities.

• Including a bash kernel in the default hub image: for the ease of using command-line tools. The default hub image would ship with two kernels (Python and Bash). Users can also get an idea of how a kernel works with the Jupyter Notebook.
• Giving users the freedom to add whatever kernel they want without modifying the hub image or root permission. It should be possible to have a kernel installed even in a shared location without being part of the image.
• Better documentation: conda-forge/pip can provide Julia, R, Python, and bash kernels, so these kernels can be readily installed on a per-user basis. With that said, it would be good to write some documentation about the how-tos and troubleshooting.