bench: survive server loss during polling (mark run failed, keep suite alive)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

bench/minimax_bench.py

@@ -196,9 +196,21 @@ def run_one(api, cfg, poll=2.0, timeout=3600):
     prompt_id = resp["prompt_id"]
     exec_start = exec_end = None
     status_str = None
+    poll_failures = 0
     while time.time() - t0 < timeout:
         time.sleep(poll)
-        h = api.history(prompt_id)
+        try:
+            h = api.history(prompt_id)
+            poll_failures = 0
+        except Exception as e:
+            poll_failures += 1
+            if poll_failures >= 10:
+                stop.set()
+                return {**cfg, "ok": False, "status": "server_lost",
+                        "wall_s": round(time.time() - t0, 1), "exec_s": None,
+                        "peak_vram_gb": round(peak["vram"] / 1e9, 1) if peak["vram"] else None,
+                        "outputs": [], "error": f"server unreachable while polling: {e}"}
+            continue
         if prompt_id in h:
             entry = h[prompt_id]
             status_str = entry.get("status", {}).get("status_str")