Save actual outputs back to notebook on disk

[ellisonbg]

Currently, in this repo, we are saving outputs to disk and writing small placeholder outputs in the ydoc. The frontend transforms these placeholder outputs into the real output by making a GET request against /api/outputs. However, we are currently saving the placeholder outputs to disk rather than the real output. This doesn't break the UX of JupyterLab, but does mean that the real outputs are not in the notebook. For some users, this may not matter, but for others it may (for example if you want to version control the notebook with outputs, or convert it to another format using nbformat, jupyter book, or myst). Users can turn this system off, but it will lead to the full outputs being stored in the ydoc again. Some options for the future:

• Re-inject the real outputs into the ydoc each time we save the notebook to disk, and replace them by the placeholders upon hydrating the ydoc.
• Provide a command line tool that allows users to generate a version of the notebook that does have the outputs.

My sense is that for most users, the current state will be a huge improvement and that we should launch with the repo in its current state and gather input from users.

[ellisonbg]

@dlqqq @3coins We will need the read portion of this logic to handle notebooks that users have created previously that have output on disk.

[minrk]

I just started looking into whether we can use this for a class, and this is a blocker for us, since it means we can't use anything other than jupyterlab to interact with notebooks, and we need actual notebooks for myst, nbval, etc.

Not doing this also breaks lots of ways of interacting with notebooks, e.g. it would appear that renaming and copying notebooks from outside jupyterlab loses the relationship between outputs and cells. Writing the outputs to the document on disk should be far more robust to the many ways folks work with notebooks.

[ellisonbg]

Fixed by #163