fix(ai-chat): 优化引用文档ID处理支持字符串数组类型

- 将数据库中 referenced_doc_ids 字段从 BIGINT[] 修改为 VARCHAR(255)[]
- 在实体类 AiChatMessage 中将 referencedDocIds 类型改为 String[] 并添加自定义类型处理器
- 新增 PostgresArrayTypeHandler 用于处理 PostgreSQL varchar 数组与 Java String[] 的映射
- 修改查询时 project_id 和 timeline_node_id 的过滤表达式，使用字符串匹配避免类型错误
- AiChatServiceImpl 中保存消息时改用字符串数组保存引用文档ID
- KbDocumentVO 新增 fileUrl 字段映射数据库中对应字段
- 数据库映射文件 AiDocumentMapper.xml 增加 file_url 字段映射

src/main/java/cn/yinlihupo/service/ai/impl/AiChatServiceImpl.java

@@ -134,9 +134,14 @@ public class AiChatServiceImpl implements AiChatService {
                     () -> {
                         // 保存助手消息
                         int responseTime = (int) (System.currentTimeMillis() - startTime);
+                        // 提取引用的文档ID列表
+                        String[] docIds = retrievedDocs.stream()
+                                .map(doc -> doc.getMetadata().getOrDefault("doc_id", "").toString())
+                                .filter(id -> !id.isEmpty())
+                                .toArray(String[]::new);
                         Long messageId = saveMessage(finalSessionId, userId, request.getProjectId(),
                                 request.getTimelineNodeId(), "assistant",
-                                fullResponse.toString(), JSON.toJSONString(referencedDocs));
+                                fullResponse.toString(), docIds);
 
                         // 发送完成消息
                         sendEvent(emitter, "complete", Map.of(
@@ -334,7 +339,7 @@ public class AiChatServiceImpl implements AiChatService {
      */
     private Long saveMessage(String sessionId, Long userId, Long projectId,
                              Long timelineNodeId, String role, String content,
-                             String referencedDocIds) {
+                             String[] referencedDocIds) {
         // 获取当前最大序号
         Integer maxIndex = chatHistoryMapper.selectMaxMessageIndex(sessionId);
         int nextIndex = (maxIndex != null ? maxIndex : 0) + 1;

src/main/java/cn/yinlihupo/service/ai/rag/RagRetriever.java

@@ -41,10 +41,11 @@ public class RagRetriever {
      */
     public List<Document> vectorSearch(String query, Long projectId, int topK) {
         try {
+            // project_id 在 metadata 中是字符串类型，需要用字符串比较
             SearchRequest searchRequest = SearchRequest.builder()
                     .query(query)
                     .topK(topK)
-                    .filterExpression("project_id == " + projectId + " && status == 'active'")
+                    .filterExpression("project_id == '" + projectId + "' && status == 'active'")
                     .build();
 
             List<Document> results = vectorStore.similaritySearch(searchRequest);
@@ -68,12 +69,13 @@ public class RagRetriever {
     public List<Document> vectorSearchWithTimeline(String query, Long projectId,
                                                     Long timelineNodeId, int topK) {
         try {
+            // project_id 和 timeline_node_id 在 metadata 中是字符串类型
             SearchRequest searchRequest = SearchRequest.builder()
                     .query(query)
                     .topK(topK)
-                    .filterExpression("project_id == " + projectId +
-                            " && timeline_node_id == " + timelineNodeId +
-                            " && status == 'active'")
+                    .filterExpression("project_id == '" + projectId +
+                            "' && timeline_node_id == '" + timelineNodeId +
+                            "' && status == 'active'")
                     .build();
 
             List<Document> results = vectorStore.similaritySearch(searchRequest);