feat(m8): P2 pgvector + Ollama 向量化与语义检索

- PostgreSQL 切换 pgvector/pgvector:pg17 镜像;迁移 000024 建 vector 扩展、
  knowledge_chunks.embedding vector(1024) + HNSW 余弦索引,retrieval_mode 放宽三态
- OllamaEmbedder 本地 bge-m3 批量嵌入,404 惰性 pull 重试,维度/超时校验,可整体关闭
- SemanticRetriever/HybridRetriever + NewRetriever 按 retrieval_mode 分发,缺 embedder 回退 FTS
- 文档入库同步批量向量化;Ollama 故障降级入库 + embedding_failed 事件
- 修复 pgx CopyFrom 对 vector 列二进制编码误读:COPY 基础列后同事务 unnest 批量回填
- 修复降级路径 embeddings=nil 索引越界 panic(Add 与 Reprocess)
- 知识库列表 vectorized_chunk_count + 前端三态检索模式选择与向量化覆盖率
- 单测 embedder/retrievers + 集成 TestKnowledgeVectorLifecycle 全绿

Co-Authored-By: Claude <noreply@anthropic.com>

internal/workbench/workbench_integration_test.go

@@ -66,7 +66,7 @@ func TestWorkbenchPostgreSQLLifecycle(t *testing.T) {
 	if doc.ChunkCount == 0 {
 		t.Fatal("expected chunks")
 	}
-	hits, err := NewPostgreSQLRetriever(assets).Search(ctx, kb.ID, "不可变运行时快照", 4)
+	hits, err := NewRetriever(assets, nil).Search(ctx, kb.ID, "不可变运行时快照", 4)
 	if err != nil || len(hits) == 0 {
 		t.Fatalf("hits=%#v err=%v", hits, err)
 	}
@@ -116,7 +116,7 @@ func TestWorkbenchPostgreSQLLifecycle(t *testing.T) {
 		}
 		_ = json.NewEncoder(w).Encode(map[string]any{"choices": []any{map[string]any{"message": map[string]any{"role": "assistant", "content": "完成"}}}})
 	})
-	runtime := NewRuntimeHTTPHandler(assets, tools, NewPostgreSQLRetriever(assets), staticPrincipalAuthenticator{}, fakeGateway, MarketplaceDeps{})
+	runtime := NewRuntimeHTTPHandler(assets, tools, NewRetriever(assets, nil), staticPrincipalAuthenticator{}, fakeGateway, MarketplaceDeps{})
 	runtimeRequest := httptest.NewRequest(http.MethodPost, "/v1/applications/m4_app/chat/completions", bytes.NewBufferString(`{"messages":[{"role":"user","content":"不可变运行时快照是什么？"}],"variables":{"question":"架构"}}`))
 	runtimeRequest.Header.Set("Authorization", "Bearer test")
 	runtimeRequest = runtimeRequest.WithContext(gateway.WithRequestID(runtimeRequest.Context(), "m4-runtime"))