feat: NanoClaude Phase 1 — 비동기 job 기반 AI Gateway 코어 구현

POST /chat → job_id ACK, GET /chat/{job_id}/stream → SSE 스트리밍,
EXAONE Ollama adapter, JobManager, StateStream, Worker 구조

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

nanoclaude/routers/chat.py

@@ -0,0 +1,61 @@
+"""Chat router — POST /chat, GET /chat/{job_id}/stream, POST /chat/{job_id}/cancel."""
+
+from __future__ import annotations
+
+import asyncio
+
+from fastapi import APIRouter, HTTPException
+from fastapi.responses import StreamingResponse
+
+from models.schemas import CancelResponse, ChatRequest, ChatResponse
+from services import worker
+from services.job_manager import job_manager
+from services.state_stream import state_stream
+
+router = APIRouter(tags=["chat"])
+
+
+@router.post("/chat", response_model=ChatResponse)
+async def create_chat(body: ChatRequest):
+    """job_id 즉시 반환 (ACK). 백그라운드에서 EXAONE 처리 시작."""
+    job = job_manager.create(body.message)
+    state_stream.create(job.id)
+
+    task = asyncio.create_task(worker.run(job))
+    job_manager.attach_task(job.id, task)
+
+    return ChatResponse(job_id=job.id)
+
+
+@router.get("/chat/{job_id}/stream")
+async def stream_chat(job_id: str):
+    """SSE 스트림으로 상태 + 결과 전달."""
+    job = job_manager.get(job_id)
+    if not job:
+        raise HTTPException(status_code=404, detail="Job not found")
+
+    return StreamingResponse(
+        _stream_with_cleanup(job_id),
+        media_type="text/event-stream",
+        headers={
+            "Cache-Control": "no-cache",
+            "X-Accel-Buffering": "no",
+        },
+    )
+
+
+async def _stream_with_cleanup(job_id: str):
+    try:
+        async for chunk in state_stream.subscribe(job_id):
+            yield chunk
+    finally:
+        state_stream.cleanup(job_id)
+
+
+@router.post("/chat/{job_id}/cancel", response_model=CancelResponse)
+async def cancel_chat(job_id: str):
+    """진행 중인 job 취소."""
+    success = job_manager.cancel(job_id)
+    if not success:
+        raise HTTPException(status_code=404, detail="Job not found or already finished")
+    return CancelResponse(status="cancelled")