fix: add preimage tags for domain separation of signatures

[cfm]

The manuscript does not specify domain-separating preimage tags. The Tamarin models, helpfully, do, so here we adopt those values.

• Closes signatures do not include message tags #159
• Closes Vulnerability found: Key Replacement on Source Submission #53

[Copilot]

Pull request overview

Updates the protocol specification to explicitly domain-separate signed messages using fixed message tags (matching the Tamarin model values).

Changes:

• Updates the SIG building block to sign/verify tag ∥ m instead of just m.
• Adds concrete domain-separation tags to signature inputs throughout the setup and key-fetch verification flows.
• Updates the signature verification checks in the “Sender fetches keys” step to include the new tags.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[cfm]

I believe this also addresses #53 by preventing type confusion between signed keys returned by the server.

[rocodes]

Just because I think there's potential for message/$message$ confusion, at least for my brain: This PR addresses binding signatures (eg selfsig over journalist long-term keys, selfsig over journalist one-time keys, newsroom sig over journalist signing key) to the pubkey_bytes ($message$) they are signing, and it doesn't have anything to do with plaintext info between parties (which are also called "messages" but which use Hpke open() and seal(), which under the hood implement their own message tag computations/checks).

Even though it's true that signed_pubkey_bytes is a "message", it might be friendly to our future selves if we try to disambiguate this a bit in our terminology when we describe this step, so that it's highly clear what it's for.

About the comment on #53: I'd say the primary mitigation was the 0.2 change in which the clue/hint dh key have no relation to anything used in message encryption. (and in 0.3, where all pubkeys that are involved encryption/decryption are either signed, enclosed in authenc ciphertext, or committed to in authenc ciphertext). But I agree that separating keys by context has to be specified.

[cfm]

That's a good disambiguation. In e46a6d5 I've suggested (and documented) s/message/preimage/g specifically for the signature scheme. Let me know what you think.

[cfm]

Rebased from main after #156.