[Experimental] JupyterLab integration

[jcheng5]

• %%reactive cell magic does not reliably clear previous output or put new output in the right place
• Create non-notebook equivalent of %%reactive
  • Display all top-level expressions (or at LEAST the last one). Also useful for express mode
  • Auto-detect if a matplotlib figure has changed and show it
  • Have displayhook render all objects like Jupyter does
  • ipywidgets? 😬
• On cell clear/delete/etc., kill any observers that originated from that cell
• Re-establish connections with existing kernels on page load (right now, connections are only opened from the kernel side, when the ipython extension loads)
• Figure out what to do when opening an .ipynb with Shiny inputs/outputs. If it doesn't "just work" make sure we let the user know they need to re-run or whatever. (Like for example the %load_ext shiny.notebook won't call itself, right?)

---

• Replace triple-click gesture with JupyterLab context menu
• Separate into separate repo, maybe
• Document everything
• Internal documentation
• Less hackery on Shiny internals (especially setting default contexts)
• Try to add Lit dependency*
• Try to add @jupyterlab/console dependency and get things working for console*
• Figure out where to log errors, especially ones caught by the Effect in the %%reactive implementation
• If session closes (on the server side), due to e.g. observer throwing unhandled exception, needs to have a visual indication for the user as well as some idea of how to proceed. OR, don't close the session on unhandled exception, but instead show the error in the owning block.

* Last time I tried to add a JS dependency, jlpm build stopped working and I had to delete the extension and start over