Deduper looks for duplicate and near duplicate photos in an Immich library. It
reads the library over the Immich API and its database, computes an embedding
for every picture with a local model and keeps the vectors in Qdrant, which
lets it group pictures that differ only in resolution, crop or compression.

The findings are presented in a web interface, where the copies to keep and
the ones to delete are picked by hand or by rules.
