feat: AI Gateway Phase 1 - FastAPI 코어 구현

GPU 서버 중앙 AI 라우팅 서비스 초기 구현:
- OpenAI 호환 API (/v1/chat/completions, /v1/models, /v1/embeddings)
- 모델 레지스트리 + 백엔드 헬스체크 (30초 루프)
- Ollama SSE 프록시 (NDJSON → OpenAI SSE 변환)
- JWT 인증 이중 경로 (httpOnly 쿠키 + Bearer 토큰)
- owner/guest 역할 분리, 로그인 rate limiting
- 백엔드별 rate limiting (NanoClaude 대비)
- SQLite 스키마 사전 정의 (aiosqlite + WAL)
- Docker Compose + Caddy 리버스 프록시

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

docker-compose.yml

@@ -0,0 +1,47 @@
+services:
+  caddy:
+    image: caddy:2-alpine
+    container_name: gpu-caddy
+    restart: unless-stopped
+    ports:
+      - "80:80"
+      - "443:443"
+    volumes:
+      - ./caddy/Caddyfile:/etc/caddy/Caddyfile
+      - caddy_data:/data
+    depends_on:
+      - hub-api
+    networks:
+      - gateway-net
+
+  hub-api:
+    build: ./hub-api
+    container_name: gpu-hub-api
+    restart: unless-stopped
+    environment:
+      - OWNER_PASSWORD=${OWNER_PASSWORD}
+      - GUEST_PASSWORD=${GUEST_PASSWORD}
+      - JWT_SECRET=${JWT_SECRET}
+      - BACKENDS_CONFIG=/app/config/backends.json
+      - CORS_ORIGINS=${CORS_ORIGINS:-http://localhost:5173}
+      - DB_PATH=/app/data/gateway.db
+    volumes:
+      - hub_data:/app/data
+      - ./backends.json:/app/config/backends.json:ro
+    extra_hosts:
+      - "host.docker.internal:host-gateway"
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
+      interval: 15s
+      timeout: 5s
+      retries: 3
+    networks:
+      - gateway-net
+
+volumes:
+  caddy_data:
+  hub_data:
+
+networks:
+  gateway-net:
+    name: gpu-gateway-network