fix(dingtalk): async reply pattern — immediate ack + batchSend for LLM response

- Send '🤔 小虾米正在思考，稍等...' immediately via sessionWebhook on each message
- Await LLM bridge call (serial queue preserved) then deliver response via batchSend
- batchSend decoupled from sessionWebhook — works regardless of webhook state
- Fix duplicate const staffId declaration (TS compile error)
- TASK_TIMEOUT_S=55 passed explicitly to bridge (was using bridge default 25s)
- senderStaffId-first routing (OAuth binding) with senderId fallback (code binding)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

packages/services/agent-service/src/infrastructure/dingtalk/dingtalk-router.service.ts

@@ -615,8 +615,16 @@ export class DingTalkRouterService implements OnModuleInit, OnModuleDestroy {
     }
 
     const bridgeUrl = `http://${instance.serverHost}:${instance.hostPort}/task`;
-    let reply: string;
 
+    // sessionWebhook TTL is ~90 minutes (per DingTalk docs), but delivering the actual LLM
+    // response synchronously makes the user wait with no feedback. Strategy:
+    //   1. Immediately send "处理中..." via sessionWebhook — user sees instant acknowledgment
+    //   2. Await the bridge call (LLM processing) — the serial queue still blocks here,
+    //      preventing concurrent LLM calls for the same user
+    //   3. Always deliver the actual response via batchSend — decoupled from webhook window
+    this.reply(msg, '🤔 小虾米正在思考，稍等...');
+
+    let reply: string;
     try {
       const result = await this.httpPostJson<{ ok: boolean; result?: unknown; error?: string }>(
         bridgeUrl,
@@ -624,7 +632,6 @@ export class DingTalkRouterService implements OnModuleInit, OnModuleDestroy {
           prompt:         text,
           sessionKey:     `agent:main:dt-${userId}`,
           idempotencyKey: msg.msgId,
-          // Pass explicit timeout to bridge — default is 25s which is too short for LLM calls.
           timeoutSeconds: TASK_TIMEOUT_S,
         },
         (TASK_TIMEOUT_S + 10) * 1000,
@@ -642,41 +649,41 @@ export class DingTalkRouterService implements OnModuleInit, OnModuleDestroy {
       reply = '与小龙虾通信时出现错误，请稍后重试。';
     }
 
-    // Try sessionWebhook first; if it has expired by the time we have a reply (LLM took
-    // longer than ~30s), fall back to proactive batchSend so the reply still reaches the user.
-    const webhookExpiry = msg.sessionWebhookExpiredTime > 1e11
-      ? msg.sessionWebhookExpiredTime
-      : msg.sessionWebhookExpiredTime * 1000;
+    await this.batchSend(staffId, reply, msg.msgId);
+  }
 
-    if (Date.now() <= webhookExpiry) {
-      this.reply(msg, reply);
-    } else {
-      this.logger.warn(
-        `sessionWebhook expired for msgId=${msg.msgId} — falling back to batchSend for userId=${userId}`,
-      );
-      const staffId = msg.senderStaffId?.trim();
-      if (staffId) {
-        this.getToken()
-          .then((token) =>
-            this.httpsPost<unknown>(
+  /** Send a proactive message to a DingTalk user via batchSend. Used for LLM replies
+   *  so that users receive the response regardless of sessionWebhook state. */
+  private batchSend(staffId: string | undefined, content: string, msgId: string): Promise<void> {
+    if (!staffId) {
+      this.logger.warn(`batchSend skipped — no staffId for msgId=${msgId}`);
+      return Promise.resolve();
+    }
+    // Chunk content to stay within DingTalk's message size limit
+    const safe   = content.replace(/\s+at\s+\S+:\d+:\d+/g, '').trim() || '（空响应）';
+    const chunks: string[] = [];
+    for (let i = 0; i < safe.length; i += DINGTALK_MAX_CHARS) {
+      chunks.push(safe.slice(i, i + DINGTALK_MAX_CHARS));
+    }
+    return this.getToken()
+      .then(async (token) => {
+        for (const chunk of chunks) {
+          await this.httpsPost<unknown>(
             'api.dingtalk.com',
             '/v1.0/robot/oToMessages/batchSend',
             {
               robotCode: this.clientId,
               userIds:   [staffId],
               msgKey:    'sampleText',
-                msgParam:  JSON.stringify({ content: reply }),
+              msgParam:  JSON.stringify({ content: chunk }),
             },
             { 'x-acs-dingtalk-access-token': token },
-            ),
-          )
-          .catch((e: Error) =>
-            this.logger.error(`batchSend fallback failed for msgId=${msg.msgId}:`, e.message),
           );
-      } else {
-        this.logger.warn(`No staffId for batchSend fallback, reply lost for msgId=${msg.msgId}`);
-      }
         }
+      })
+      .catch((e: Error) =>
+        this.logger.error(`batchSend failed for msgId=${msgId}:`, e.message),
+      );
   }
 
   // ── Reply (chunked) ────────────────────────────────────────────────────────