refactor(chat): marshal via the framework serializer, not hand-rolled JSON

[nickmeinhold]

What & why

Andy's note on #4 named Aiko's core principle: developers design at the level of function calls and let the framework insert serialization + transport — they shouldn't hand-write the wire format. chat.py was doing exactly that with json.dumps / json.loads.

This swaps the hand-rolled marshalling for the framework's pluggable serializer (aiko_services.main.utilities.generate / parse) — the same machinery the discovery proxy already uses for remote calls.

Changes

• generate_payload() → generate("message", {username, channel, timestamp, message}). Emits an S-expression today; swappable to JSON/AVRO without touching this code.
• format_incoming() → decodes via parse(), falling back to legacy JSON then a bare string, so older publishers keep working.

Verified (round-trip + both fallbacks)

• S-expr (message username: nick … message: 11:hello world) → nick: hello world
• legacy JSON {"username":"deanna","message":"hi"} → deanna: hi
• bare string just text → just text

⚠️ Wire-format change — needs a follow-up

Published messages move from JSON to S-expressions. Reads here are backward-compatible, but external consumers that parse the JSON directly — notably the aiko-bridge — need a matching parse() update. Flagging for sequencing; happy to land the bridge change alongside.

Addresses Andy's #4 comment / the "let the framework do the marshalling" migration.

🤖 Generated with Claude Code

[nickmeinhold]

Cage-match + companion bridge PR

Run through adversarial review — Maxwell (Claude), Kelvin (Gemini 2.5 Pro), Carnot (Codex). Verdicts: 2× REQUEST_CHANGES + 1× COMMENT. Consensus findings, all fixed in ed1608d:

• Narrowed except Exception → (ValueError, IndexError, TypeError), so a real parser bug surfaces instead of silently degrading to the JSON/raw fallback.
• Hoisted the magic command string "message" to a module constant _MESSAGE_COMMAND (used by both encode + decode).
• Required the message field in the S-expr branch (Carnot's catch) so a malformed (message username: nick) falls through to raw instead of rendering an empty nick: .

Wire-format follow-up — resolved: the consumer update for the aiko-bridge is up as nickmeinhold/aiko-bridge#1. It reads S-expr → legacy JSON → bare string, so it works against both old and new servers (no flag-day). The two land together.

[nickmeinhold]

Noting a follow-up from the protocol version bump (dea848a, :0 -> :1) for when this lands:

The aiko-bridge discovers the ChatServer via get_server_service_filter(), which pins the protocol version (_PROTOCOL_SERVER = .../chat_server:{_VERSION}, currently _VERSION = 0). Once the deployed server moves to :1, the bridge needs to track that. And since the bridge genuinely depends on the new functionality (it uses username to mint per-user Matrix ghosts), per your discover-by-version guidance it should arguably require :1 rather than accept either, so it never silently attaches to a :0 server that can't carry sender identity.

Tracking it bridge-side; not a change to this PR. Flagging here since it's the same wire-format thread. (Same direction as this PR, incidentally: the robot half of the bridge just moved off hand-rolled marshaling onto the framework's remote-call proxy, so commands are function calls and the framework handles serialization.)