Cache ZIM metadata on library-gen?

Currently, the library generator script which is used both for library and dev-library (different source folders) spends most of its time reading metadata from ZIM files on the filesystem.

On library, this is ~6,800 files. This can be completed within ~6mn but if the disk is busy (reminder: the server is using mechanical drives), this can take 3 hours.

This script is ran every 30mn on library and every 10mn for dev-library.

While this will all be obsolete once the CMS takes over, a quick and easy improvement would be to cache this information and only read metadata for new files. It's actually already cached (in previously written library xml) so it's just a matter of skipping/reusing data for existing entries.

The only drawback is that it wont update metadata of a file that has been overwritten but that's already a scenario we've excluded and we could implement a simple file-flag that triggers a full re-read if present.

Start by locating the Python library generator used for both library and dev-library, then inspect how it reads ZIM metadata and writes the existing library XML. Reuse metadata for entries already present and read metadata only for new files; verify repeated runs avoid unnecessary filesystem reads while preserving the existing generation behavior.

Written by the indexing model from the issue text.