fix: 取消blocked，改为needs_llm升级机制

- executor重试耗尽→needs_llm(需知微介入)，非blocked
- health check报告尾段展示needs_llm项+失败原因
- derive_fix_action覆盖全部已知场景(cron errors/delivery等)
- TODO创建时注明"无法当场修复原因"
- 每个TODO必有fix_action，没有的不创建TODOs直接在报告列出

scripts/self_todo_executor.py

@@ -88,12 +88,12 @@ def main():
         else:
             retry_count += 1
             if retry_count >= max_retries:
-                # 重试用完，留到下次体检再重新发现
+                # 重试用完，超出能力范围→标记需LLM处理
                 conn.execute(
-                    "UPDATE todos SET status='pending', retry_count=0, "
+                    "UPDATE todos SET status='needs_llm', retry_count=?, "
                     "note=?, updated_at=CURRENT_TIMESTAMP WHERE id=?",
-                    (f"已达最大重试({max_retries}次)，留待下次", todo_id))
-                results.append(("🔄", f"{title}: 重试耗尽，等下次体检"))
+                    (retry_count, f"尝试{retry_count}次仍失败: {output[:150]}", todo_id))
+                results.append(("🔶", f"{title}: 需知微介入(重试{retry_count}次失败)"))
             else:
                 conn.execute(
                     "UPDATE todos SET status='pending', retry_count=?, "