feat: QYLAW 法律法规知识库

- 语义检索(FAISS + embedding)+ 精确查找(法规名+条号)
- RAG 问答(SSE 流式,支持 thinking 折叠显示)
- 法规浏览(原文阅读)
- 历史记录(检索+对话持久化到 SQLite)
- 设置页(系统提示词/模板/LLM 参数可配置)
- 检索质量评估脚本

Generated with [Devin](https://devin.ai)

Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>

app/services/embedding.py

@@ -0,0 +1,79 @@
+"""embedding 服务客户端 — 调用 114:8003 Qwen3-Embedding 服务"""
+import asyncio
+import logging
+from typing import List
+
+import httpx
+
+from app.config import EMBEDDING_URL, EMBEDDING_MODEL, EMBEDDING_BATCH_SIZE
+
+logger = logging.getLogger(__name__)
+
+_client: httpx.AsyncClient | None = None
+
+
+def _get_client() -> httpx.AsyncClient:
+    """复用 httpx 客户端"""
+    global _client
+    if _client is None or _client.is_closed:
+        _client = httpx.AsyncClient(timeout=60)
+    return _client
+
+
+async def embed_batch(texts: List[str], batch_size: int = EMBEDDING_BATCH_SIZE) -> List[List[float]]:
+    """批量向量化
+
+    Args:
+        texts: 文本列表
+        batch_size: 每批大小,默认 32
+
+    Returns:
+        向量列表,顺序与输入一致
+
+    Raises:
+        RuntimeError: embedding 服务不可用或返回错误
+    """
+    if not texts:
+        return []
+
+    client = _get_client()
+    all_vecs: List[List[float]] = []
+
+    for i in range(0, len(texts), batch_size):
+        batch = texts[i : i + batch_size]
+        payload = {
+            "model": EMBEDDING_MODEL,
+            "input": batch,
+        }
+
+        # 重试 3 次,指数退避
+        last_err = None
+        for attempt in range(3):
+            try:
+                resp = await client.post(
+                    f"{EMBEDDING_URL}/embeddings",
+                    json=payload,
+                )
+                resp.raise_for_status()
+                data = resp.json()
+                # vLLM embeddings 响应格式: {data: [{embedding: [...]}], ...}
+                vecs = [item["embedding"] for item in data["data"]]
+                all_vecs.extend(vecs)
+                last_err = None
+                break
+            except Exception as e:
+                last_err = e
+                wait = 2 ** attempt
+                logger.warning(f"embedding 批次 {i//batch_size} 第 {attempt+1} 次失败: {e}, {wait}s 后重试")
+                await asyncio.sleep(wait)
+
+        if last_err is not None:
+            raise RuntimeError(f"embedding 服务调用失败(重试 3 次): {last_err}")
+
+    return all_vecs
+
+
+async def embed_text(text: str) -> List[float]:
+    """单条文本向量化"""
+    vecs = await embed_batch([text])
+    return vecs[0]