fix: per_stock/self_repair按文档去掉显式max_tokens(改None不指定)

deploy/profile-scripts/per_stock_reassess.py

@@ -405,7 +405,7 @@ def main():
                     if _prompt:
                         try:
                             # 2026-08-13 超时修复：timeout 150→90, retries 1→0（3只×(90+90+20)=600s临界,减到90+0重试=270s安全）
-                            _llm_result = call_llm(_prompt, max_tokens=4096, timeout=90, retries=0, backoff=0)
+                            _llm_result = call_llm(_prompt, max_tokens=None, timeout=90, retries=0, backoff=0)  # 文档: 不指定max_tokens
                             if _llm_result["ok"]:
                                 _full_analysis_text = _llm_result["content"]
                                 print(f"  ✅ LLM12维分析完成({len(_full_analysis_text)}字, {_llm_result['elapsed']:.1f}s)", flush=True)

deploy/profile-scripts/self_repair.py

@@ -179,7 +179,7 @@ def llm_diagnose(failure):
         # session 指令冻结+上下文无限累积，2026-07-21 废弃）
         sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
         from llm_client import call_llm
-        res = call_llm(prompt, max_tokens=300, timeout=90, retries=0, backoff=0, concurrent=True)
+        res = call_llm(prompt, max_tokens=None, timeout=90, retries=0, backoff=0, concurrent=True)  # 文档: 不指定max_tokens
         if not res["ok"]:
             raise RuntimeError(res["error"][:120])
         text = res["content"]