serve: ACCEPT frame — oversized prompt is a clean 400, not a broken 200 stream (#597 item 6)

[JustVugg]

Fixes #597 item 6 — @mkelcb's diagnosis and protocol design.

The bug

A streaming chat request committed its HTTP 200 (SSE headers) before the engine tokenised the prompt and checked it against CTX. An oversized prompt was only discovered afterwards (ERROR CONTEXT_EXCEEDED) — too late to send the 400 the client needs, so the client got a broken/empty stream instead of a clean context_length_exceeded. (Non-streaming was already correct: it sends 200 only after generation completes.)

The fix — an acceptance frame in the engine↔gateway protocol

• c/colibri.c mux_submit() — after all cheap validation (encode, non-empty, fits CTX) and before prefill, emit ACCEPT <id> <ntok>. Every early failure already returns first with an ERROR, so a request yields exactly one of ACCEPT or an early ERROR.
• Engine._dispatch_stdout() — parse ACCEPT → ("accept", …) without popping the pending request (DATA/DONE still follow).
• Engine.generate() — new on_accept callback; commit exactly once on the first of ACCEPT / DATA / DONE. DATA/DONE before an ACCEPT still commit, so an older engine with no ACCEPT frame keeps working; a second ACCEPT is a protocol error.
• APIHandler.generation() streaming — defer the SSE 200 + keepalive into start_stream(), fired by on_accept. An ERROR before ACCEPT now propagates as an APIError with nothing committed → do_POST returns a normal JSON 400.

Tests

4 Engine-level (fake engine process): ACCEPT-before-DATA fires on_accept first; ERROR-before-ACCEPT raises 400 and never commits; DATA-before-ACCEPT still commits (old engine); duplicate ACCEPT is a protocol error.
1 end-to-end HTTP: a streaming context-exceeded request returns a clean 400 context_length_exceeded, not a 200 stream.
Full server suite 83/83; the C engine builds clean.

#597 is now complete

#	item	status
1	<|user|> turn termination	✅ #535
2	reverse-proxy Host allowlist	✅ #606
3	HTTP/1.1 desync	mitigated on dev — awaiting reporter retest
4	reasoning via reasoning_content	✅ #607
5	quiet prefill keepalive	✅ #606
6	CONTEXT_EXCEEDED after 200	✅ this PR