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/platform/config/config.go

@@ -22,8 +22,9 @@ type Config struct {
 	Audit        Audit
 	Outbox       Outbox
 	RuntimeData  RuntimeData
-	Shadow       Shadow
+	Shadow        Shadow
 	ObjectStorage ObjectStorage
+	Embeddings    Embeddings
 }
 
 type Server struct {
@@ -125,6 +126,17 @@ type ObjectStorage struct {
 	MaxFileBytes    int64
 }
 
+// Embeddings 配置本地 Ollama 向量化(默认 bge-m3)。Enabled 为 false 时网关不构造
+// OllamaEmbedder,知识库退回纯 FTS 检索,AddKnowledgeDocument 不生成 embedding。
+type Embeddings struct {
+	Enabled   bool
+	BaseURL   string
+	Model     string
+	Dim       int
+	BatchSize int
+	Timeout   time.Duration
+}
+
 func Load() (Config, error) {
 	cfg := Config{
 		Environment: env("APP_ENV", "local"),
@@ -202,6 +214,14 @@ func Load() (Config, error) {
 			UseSSL:          boolValue("S3_USE_SSL", false),
 			MaxFileBytes:    int64Value("S3_MAX_FILE_BYTES", 128<<20),
 		},
+		Embeddings: Embeddings{
+			Enabled:   boolValue("EMBEDDINGS_ENABLED", true),
+			BaseURL:   strings.TrimRight(env("OLLAMA_BASE_URL", "http://ollama:11434"), "/"),
+			Model:     env("EMBEDDING_MODEL", "bge-m3"),
+			Dim:       intValue("EMBEDDING_DIM", 1024),
+			BatchSize: intValue("EMBEDDING_BATCH_SIZE", 64),
+			Timeout:   duration("EMBEDDING_TIMEOUT", 120*time.Second),
+		},
 	}
 
 	return cfg, cfg.Validate()
@@ -271,6 +291,27 @@ func (c Config) Validate() error {
 	if c.ObjectStorage.MaxFileBytes < 1<<20 || c.ObjectStorage.MaxFileBytes > 512<<20 {
 		errs = append(errs, errors.New("S3_MAX_FILE_BYTES must be between 1 MiB and 512 MiB"))
 	}
+	if c.Embeddings.Enabled {
+		if err := validateHTTPURL(c.Embeddings.BaseURL); err != nil {
+			errs = append(errs, fmt.Errorf("OLLAMA_BASE_URL: %w", err))
+		}
+		if c.Embeddings.Model == "" {
+			errs = append(errs, errors.New("EMBEDDING_MODEL is required when embeddings are enabled"))
+		}
+		if c.Embeddings.Dim < 128 || c.Embeddings.Dim > 8192 {
+			errs = append(errs, errors.New("EMBEDDING_DIM must be between 128 and 8192"))
+		}
+		if c.Embeddings.Dim != 1024 {
+			// 知识库 embedding 列固定为 vector(1024);维度不符会让入库向量报错。
+			errs = append(errs, errors.New("EMBEDDING_DIM must be 1024 to match the vector(1024) column"))
+		}
+		if c.Embeddings.BatchSize < 1 || c.Embeddings.BatchSize > 512 {
+			errs = append(errs, errors.New("EMBEDDING_BATCH_SIZE must be between 1 and 512"))
+		}
+		if c.Embeddings.Timeout < time.Second || c.Embeddings.Timeout > 30*time.Minute {
+			errs = append(errs, errors.New("EMBEDDING_TIMEOUT must be between 1s and 30m"))
+		}
+	}
 	return errors.Join(errs...)
 }