feat(rag): dense+BM25 hybrid on MilvusClientV2, drop SDK path

Replace legacy MilvusServiceClient knowledge search/write with a single
MilvusClientV2 hybrid store (BM25 function + dense ANN + RRFRanker).
Use collection biz_hybrid and require knowledge reindex.

src/main/resources/application.yml

@@ -25,6 +25,8 @@ milvus:
   token: ${MILVUS_TOKEN}
   secure: true
   vector-dim: 1024  # BGE-M3 = 1024，换模型时同步改
+  # dense+BM25 hybrid collection (requires reindex; legacy "biz" is not used by knowledge path)
+  collection: biz_hybrid
 
 # =====================================================
 # 模型路由配置
@@ -162,16 +164,24 @@ rag:
       enabled: false
       content-preview-limit: 300
 
-# 检索归一化配置
+# 检索配置（单一 Milvus V2 后端；已移除 sdk/spring/auto 路由）
 retrieval:
-  kb-scope: ""                         # empty means search all legacy documents; use rag-eval for eval seed docs
-  vector-store:
-    mode: auto # auto | spring-ai | sdk
+  kb-scope: ""                         # empty means search all documents in hybrid collection
+  search:
+    mode: hybrid                       # dense | hybrid (dense + BM25 RRF)
+  hybrid:
+    rrf-k: 60
   normalization:
     max-l2-distance: 2.0              # L2 距离上界（BGE-M3 单位向量 = 2.0）
     highly-relevant-threshold: 0.75   # similarity >= 0.75 → HIGHLY_RELEVANT
     reference-threshold: 0.5          # similarity >= 0.5 → REFERENCE
 
+# RAG 召回宽度（Delivery 1）
+rag:
+  retrieve-k: 20
+  return-n: 5
+  max-chunks-per-document: 2
+
 # Prometheus 配置
 prometheus:
   base-url: http://localhost:9090