[MISC] Cherry pick #1291 from v0.9.1-dev (#1825)

### What this PR does / why we need it?
Cherry pick #1291 from v0.9.1-dev, This pr implement the synchronization
of whether `dbo` is enabled across all dp ranks. specifically, it
performed allreduce op across multiple DP ranks, only when all the dp
rank is `enable_dbo`, it is enabled

Co-authored-by: shikang-hangzhou <459956190@qq.com>
Co-authored-by: wangli <wangli858794774@gmail.com>

- vLLM version: v0.10.0
- vLLM main:
2836dd73f1

---------

Signed-off-by: wangli <wangli858794774@gmail.com>

vllm_ascend/attention/attention_v1_torchair.py

@@ -140,6 +140,8 @@ class AscendTorchairMetadata:
 
     decode: Optional[AscendDecodeMetadata] = None
 
+    enable_dbo_across_dp: bool = False
+
 
 class AscendAttentionTorchairMetadataBuilder:
 
@@ -220,7 +222,8 @@ class AscendAttentionTorchairMetadataBuilder:
               num_reqs,
               num_actual_tokens,
               max_query_len,
-              graph_pad_size: int = -1):
+              graph_pad_size: int = -1,
+              enable_dbo_across_dp: bool = False):
 
         device = self.runner.device
 
@@ -298,7 +301,8 @@ class AscendAttentionTorchairMetadataBuilder:
             max_query_len=max_query_len,
             slot_mapping=slot_mapping,
             attn_mask=attn_mask,
-            attn_state=attn_state)
+            attn_state=attn_state,
+            enable_dbo_across_dp=enable_dbo_across_dp)
         return attn_metadata