fix(story-summary): improve L0 cancellation and tuning

modules/story-summary/vector/llm/atom-extraction.js

@@ -8,14 +8,14 @@
 // 每楼层 1-2 个场景锚点（非碎片原子），60-100 字场景摘要
 // ============================================================================
 
-import { callLLM, parseJson } from './llm-service.js';
+import { callLLM, cancelAllL0Requests, parseJson } from './llm-service.js';
 import { xbLog } from '../../../../core/debug-core.js';
 import { filterText } from '../utils/text-filter.js';
 
 const MODULE_ID = 'atom-extraction';
 
 const CONCURRENCY = 10;
-const RETRY_COUNT = 2;
+const RETRY_COUNT = 1;
 const RETRY_DELAY = 500;
 const DEFAULT_TIMEOUT = 40000;
 const STAGGER_DELAY = 80;
@@ -25,6 +25,7 @@ let batchCancelled = false;
 
 export function cancelBatchExtraction() {
     batchCancelled = true;
+    cancelAllL0Requests();
 }
 
 export function isBatchCancelled() {

modules/story-summary/vector/llm/llm-service.js

@@ -9,6 +9,7 @@ const SILICONFLOW_API_URL = 'https://api.siliconflow.cn/v1';
 const DEFAULT_L0_MODEL = 'Qwen/Qwen3-8B';
 
 let callCounter = 0;
+const activeL0SessionIds = new Set();
 
 function getStreamingModule() {
     const mod = window.xiaobaixStreamingGeneration;
@@ -69,6 +70,7 @@ export async function callLLM(messages, options = {}) {
     }
 
     try {
+        activeL0SessionIds.add(uniqueId);
         const timeoutPromise = new Promise((_, reject) => {
             setTimeout(() => reject(new Error(`L0 request timeout after ${timeout}ms`)), timeout);
         });
@@ -80,9 +82,20 @@ export async function callLLM(messages, options = {}) {
     } catch (e) {
         xbLog.error(MODULE_ID, 'LLM调用失败', e);
         throw e;
+    } finally {
+        activeL0SessionIds.delete(uniqueId);
     }
 }
 
+export function cancelAllL0Requests() {
+    const mod = getStreamingModule();
+    if (!mod?.cancel) return;
+    for (const sessionId of activeL0SessionIds) {
+        try { mod.cancel(sessionId); } catch {}
+    }
+    activeL0SessionIds.clear();
+}
+
 export function parseJson(text) {
     if (!text) return null;
     let s = text.trim().replace(/^```(?:json)?\s*/i, '').replace(/\s*```$/i, '').trim();

modules/story-summary/vector/pipeline/state-integration.js

@@ -29,7 +29,7 @@ import { filterText } from '../utils/text-filter.js';
 const MODULE_ID = 'state-integration';
 
 // ★ 并发配置
-const CONCURRENCY = 50;
+const CONCURRENCY = 10;
 const STAGGER_DELAY = 15;
 const DEBUG_CONCURRENCY = true;
 const R_AGG_MAX_CHARS = 256;