Reading PACKAGES.gz over HTTPS from inside R sends an R User-Agent, which the
edge routes to a per-minor index. The script then enumerated union records
carrying `Path: <minor>`, whose objects do not exist at the flat path, so every
download 404'd: the first run skipped 292 of 300.
The bucket cannot be routed and is authoritative about which objects are
actually in the flat slot, so list and fetch from there.