chore: 清理违反底座规则的死代码并修正接口文档编号

- 删除生产死代码 app/service/knowledge_tool_service.py 与
  app/infrastructure/milvus_knowledge_adapter.py：后者硬编码 Milvus 字段名，
  违反 AGENTS.md §E，且仅被前者引用；生产检索链路实际走
  knowledge_search_tool -> KnowledgeSearchService -> knowledge_schema 运行时探测。
- 删除上述两模块的单测，以及依赖 legacy 位置参数构造的
  tests/unit/service/test_knowledge_retrieval.py。
- app/service/knowledge_retrieval_service.py 整文件回退底座版本，
  移除 legacy 双构造与重复检索实现。
- docs/05-接口文档.md：客服画像候选改登记为 §8.5，恢复 §8.2 解析知识引用；
  既有 §8.1-§8.4 编号全部保持，修复此前出现两个 8.3 的问题。
- app/model/profile.py：current_customer_id 改为普通可空列映射，与
  alembic/baseline_generated.sql 及真实库一致；原 Computed 声明会让 ORM 把该列
  从 INSERT 中排除，与「必须显式写入」的实际 schema 不符。
- 新增 docs/客服Agent接入底座扩展说明_v1.md，供集成分支评审逐项确认。

验证：pytest tests/unit tests/contract -> 1275 passed, 2 skipped, 0 failed；
ruff check app tests tools alembic 通过；mypy app 通过（244 个源文件）。

app/infrastructure/milvus_knowledge_adapter.py

@@ -1,74 +0,0 @@
-from typing import Any
-
-from app.core.errors import ForbiddenAgentError, RecoverableAgentError
-from app.core.knowledge_contracts import ALLOWED_KNOWLEDGE_COLLECTIONS
-
-
-class MilvusKnowledgeClient:
-    def __init__(self, uri: str, token: str | None = None) -> None:
-        self._uri = uri
-        self._token = token
-        self._client: Any | None = None
-
-    async def _ensure_client(self) -> Any:
-        if self._client is None:
-            from pymilvus import AsyncMilvusClient  # type: ignore[import-untyped]
-
-            self._client = AsyncMilvusClient(uri=self._uri, token=self._token)
-        return self._client
-
-    async def search(
-        self, collection: str, vector: list[float], top_k: int
-    ) -> list[dict[str, Any]]:
-        if collection not in ALLOWED_KNOWLEDGE_COLLECTIONS:
-            raise ForbiddenAgentError("未授权的知识集合")
-        if len(vector) != 1024 or not 1 <= top_k <= 20:
-            raise RecoverableAgentError("知识检索参数无效")
-        try:
-            client = await self._ensure_client()
-            # Lite 重启后集合默认未加载；远程 Milvus 对重复加载保持幂等。
-            load_collection = getattr(client, "load_collection", None)
-            if load_collection is not None:
-                await load_collection(collection_name=collection)
-            batches = await client.search(
-                collection_name=collection,
-                data=[vector],
-                limit=top_k,
-                output_fields=["knowledge_id", "title", "snippet", "tags", "version"],
-                search_params={"metric_type": "COSINE"},
-            )
-        except Exception as exc:
-            raise RecoverableAgentError("知识检索不可用") from exc
-        return [
-            normalized
-            for batch in batches
-            for hit in batch
-            if (normalized := self._normalize_hit(hit)) is not None
-        ]
-
-    @staticmethod
-    def _normalize_hit(hit: Any) -> dict[str, Any] | None:
-        """统一 Milvus SDK 的平铺与 entity 包装命中格式。"""
-        raw = dict(hit)
-        entity = raw.get("entity")
-        fields = entity if isinstance(entity, dict) else raw
-        knowledge_id = fields.get("knowledge_id")
-        snippet = fields.get("snippet")
-        score = raw.get("score", raw.get("distance", fields.get("score")))
-        if (
-            not isinstance(knowledge_id, str)
-            or not isinstance(snippet, str)
-            or not isinstance(score, (int, float))
-            or isinstance(score, bool)
-        ):
-            return None
-        normalized: dict[str, Any] = {
-            "knowledge_id": knowledge_id,
-            "snippet": snippet,
-            "score": float(score),
-        }
-        for field in ("title", "tags", "version"):
-            value = fields.get(field)
-            if value is not None:
-                normalized[field] = value
-        return normalized