patrick
ead43b2562
The frontend built book_ids with `set` rather than `append`, so the archive only ever held one book. The proxy also sent an empty Content-Length, which is not valid HTTP/1.1, and buffered the whole zip before forwarding a byte of it. Entries now carry the file's own name, suffixed when two books claim the same one, and the books are resolved before the stream starts so a miss is a 404 instead of a truncated archive under a 200.