Address comment: revert type change for deepep_ll for now

Author: minosfuture

vllm/model_executor/layers/fused_moe/deepep_ll_prepare_finalize.py

@@ -65,7 +65,7 @@ def max_num_tokens_per_rank(self) -> Optional[int]:
         return self.max_tokens_per_rank
 
     def topk_indices_dtype(self) -> Optional[torch.dtype]:
-        return torch.int32
+        return torch.int64
 
     def _do_quant(
         self,