Binder doesn't work properly

[basnijholt]

Because the notebooks are in Markdown, the default Jupyter notebook cannot read them.

[akhmerov]

There are several ways to fix that:

• Configure jupytext in binder, and use it as a default viewer for markdown files. This is much less work, but will mean that the notebooks have no pre-evaluated outputs.
• Provide a binder dockerfile that runs the conversion and serves the notebooks. This pathway would potentially lead to problems with binder image cache expiration (i.e. if nobody runs the binder for a while, the first startup will require some 30 minute wait). That could also be avoided by periodically calling binder API from CI.

[akhmerov]

The first option works in classic notebook, seems not yet possible in lab mwouts/jupytext#271. For that reason I'm going to keep the binder interface to classic for now.