[AlxcBTW]

Board task T57 (taken over from TideOPL). Bridge protocol version 1: hello no longer carries the token.

What changes on the wire

• hello is version: u16 + a 32-byte nonce. New challenge (0x0004: version, service nonce, service proof) and auth (0x0005: bridge proof). Then welcome or reject.
• A proof is HMAC-SHA-256 keyed with the token over AnyPlay bridge v1, a role byte (1 bridge, 2 service) and both nonces. The bridge checks the service's proof before it sends its own. Constant-time compare. Test vectors in the spec, checked by both sides (same bytes).
• Until the handshake is done frames are limited to 1024 bytes of payload, and the whole handshake has 5 seconds. This builds on feat(net): cap payload accepted before bridge hello #111; it extends the cap from hello to all three handshake frames.
• Version: clean bump 0 to 1. Old bridge, new service: reject naming both versions ("update AnyPS5 and the launcher"). New bridge, old service: the old service reads the nonce as a token and rejects with "token does not match"; it learns only a random nonce. No negotiation down, since version 0 is the raw token.
• ADR 0039, spec, CONTEXT.md term, crates/bridge-protocol (auth module with SHA-256/HMAC, no new dependency, so no Cargo.lock change beyond fake-game using rand), net listener runs the handshake itself (the service only sees authenticated bridges), fake-game does the real handshake, new fuzz target bridge_handshake.

Fork

The bridge change is anyplaynet/AnyPS5#31. Do not merge this before that PR's Windows build is green; the launcher's network.ref must not move to a net service whose matching AnyPS5 build is not offered.

Tested

Not built locally (no cargo on this PC today); CI is the first run. The auth primitives and vectors were cross-checked with Node's crypto and a g++ build of the C++ header.

---

^{Need help on this PR? Tag @codesmith-bot with what you need.}