ZIP preview should not write to disk

Currently the preview of a ZIP file unzips the file into a temporary cache.

This can become extremely costly in time and disk space for multi-gigabyte ZIPs, especially since most of the time the user will only see the index.html of the preview and never click on the links to the embedded files.

The index.html that we generate dynamically (when needed) can live in memory, and the other blobs can be fetched from the ZIP on demand.

—> deprecate and stop using ZipCachableBlobHolder