Wishlist: compressed notebook format or notebook format with compressed vector images

Unaware if this has been discussed already (I tend to think so, but I could not find any reference here). If so, please take my apology in advance for the noise and please be so kind to provide a pointer.

I wonder if it would make sense to have a *compressed* version of the notebook format (e.g. an `.ipynbz` format) with some official blessing. As a parallel, you could consider `svg` and `svgz` for images, where the latter is the former passed through gzip.

Unless there are strong contraindications for that, I believe that there could be the following advantages:

- being mostly text, `ipynb` files compress very well. Their gzip version is often down to 1/4 of the original file;
- when ipynb files contain svg images, these also compress extremely well.

An alternative (possibly better), would be a zip-based format (a la opendocument, docx) say a `.ipnbx` where images get stored as separate entities and the main json data only contains references to them.

A further alternative would be to at least store vector images (svg) in the ipynb file as svgz using the same base-64 encoding employed for other binary formats (e.g, png). This still would save a lot of space for notebooks containing a lot of plots in vector format. Assuming that svgz reduces the files to 25% and that base 64 re-adds 33%, this would still be just 33% of the original file size.

Start by reviewing the existing notebook format and how it represents embedded images; the issue names no files or tests. Compare the proposed compressed notebook, zip-based, and compressed-SVG alternatives with the current format, then determine whether the project wants to adopt one and document or implement that decision.

Written by the indexing model from the issue text.