debug LM caption: log result keys, check nested result, increase fetch timeout to 120s

app.py

@@ -152,14 +152,21 @@ def _caption_via_understand(audio_path, timeout=600, cancel_check=None):
         return None
 
     try:
-        r = _fetch_result(job_id, timeout=30)
+        r = _fetch_result(job_id, timeout=120)
         if r.status_code != 200:
+            logger.warning("[Caption] %s: result fetch HTTP %d", fname, r.status_code)
             return None
         data = r.json()
-        if isinstance(data, dict) and data.get("caption"):
-            return data
+        logger.info("[Caption] %s: result keys=%s", fname, list(data.keys()) if isinstance(data, dict) else type(data).__name__)
+        if isinstance(data, dict):
+            if data.get("caption"):
+                return data
+            if data.get("result") and isinstance(data["result"], dict):
+                return data["result"]
+            logger.warning("[Caption] %s: no caption in result: %s", fname, str(data)[:200])
         return None
-    except Exception:
+    except Exception as exc:
+        logger.warning("[Caption] %s: result parse failed: %s", fname, exc)
         return None