• Execution creates a JS output
• IPynb now has JS output.
• Renderer generates an image output
• Renderer then adds this image output to the list of outputs
• IPynb saved now has JS + image output.

Upon re-loading the notebook, Jupyter ends up loading the image.

Pros:

• This can be shared and image is loaded and displayed, even in github
• This is how jupyter (notebook/lab) works today

Cons

• If we changed to this behaviour, users loose out on the interactive nature of the plots.

Note:
Once #11218 is fixed, we don't need to do this, however this ensures the file can be opened on github or the like.
Also this ensures we're consistent with Jupyter.