Fix --moe-a2a-backend silently ignored for LongCat-2.0 (moe_topk missing from gate) (#30975)

Co-authored-by: whn09 <whn09@users.noreply.github.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

python/sglang/srt/managers/scheduler.py

@@ -729,12 +729,14 @@ class Scheduler(
         )
 
         # Different MoE architectures expose the per-token expert count under
-        # different attribute names (e.g. Gemma4 uses ``top_k_experts``).
+        # different attribute names (e.g. Gemma4 uses ``top_k_experts``,
+        # LongCat-2.0 uses ``moe_topk``).
         moe_topk_attrs = (
             "num_experts_per_tok",
             "num_experts_per_token",
             "top_k_experts",
             "moe_top_k",
+            "moe_topk",
         )
         if any(hasattr(config_to_check, attr) for attr in moe_topk_attrs):
             initialize_moe_config(self.server_args)