fix(pipeline): cap LLM brief size on report regen; enrich chat HTTP errors

Regenerate-report (llm) used compact_brief_for_llm default 350k chars,
unlike other LLM steps (80k–100k), which can cause gateway timeouts or 5xx
surfacing as 502. chat_completion_text now attaches response body snippet
to HTTPError for easier diagnosis.

Made-with: Cursor

backend/crawler_copy/jd_pc_search/AI_crawler.py

@@ -281,7 +281,19 @@ def chat_completion_text(
         json=body,
         timeout=timeout,
     )
-    r.raise_for_status()
+    try:
+        r.raise_for_status()
+    except requests.HTTPError as e:
+        snippet = ""
+        if e.response is not None:
+            snippet = (e.response.text or "")[:1200].replace("\r\n", "\n").replace("\n", " ")
+        if snippet:
+            raise requests.HTTPError(
+                f"{e!s} | body: {snippet}",
+                response=e.response,
+                request=e.request,
+            ) from e
+        raise
     data = r.json()
     msg = (data.get("choices") or [{}])[0].get("message") or {}
     return _normalize_chat_content(msg.get("content"))

backend/pipeline/llm_generate.py

@@ -60,7 +60,8 @@ REPORT_USER_PREFIX = """请根据以下 JSON 撰写上文所述 §8.5 嵌入段
 
 
 def generate_competitor_report_markdown_llm(brief: dict[str, Any], keyword: str) -> str:
-    compact = compact_brief_for_llm(brief)
+    # 与章节衔接等 LLM 步骤一致控制体积；默认 350k 易触发网关超时/拒收导致 502
+    compact = compact_brief_for_llm(brief, max_chars=100_000)
     payload = {
         "keyword": keyword,
         "competitor_brief": compact,