Unpack and deduplicate a matter's email

Two data structures do the work. unpack() turns one message into a flat list of leaf attachments, each with a path that says where it sat ("1/lease docs.zip/Lease 1987.pdf", "1/forwarded/1"). build_index() groups those by SHA-256 into documents, and each document keeps every place it arrived instead of the first one. The decision that matters is to hash the attachment bytes, not the message: the same email serialised twice has different MIME boundaries, so the .eml bytes differ, while the PDF inside is byte-for-byte identical.