【main】SP For Qwen3 MoE (#2209)

### What this PR does / why we need it?
Qwen3 MoE supports SP. In scenarios like AlltoAll, AlltoAllv, and MC2,
replacing AllReduce with Reduce-Scatter and AllGather achieves
computational benefits in norm operations while saving one AllGather
communication. This feature is enabled during the P-phase and delivers
notable gains in long-sequence scenarios (e.g., 16k–25k), with
performance improvements reaching 5%–10%.
### Does this PR introduce _any_ user-facing change?

### How was this patch tested?
``` 
compilation_config={
    "pass_config":{
        "enable_sequence_parallelism": True
    }
},
enable_expert_parallel=True,
```

- vLLM version: v0.10.0
- vLLM main:
9edd1db02b

---------

Signed-off-by: libaokui <libaokui@huawei.com>
Co-authored-by: libaokui <libaokui@huawei.com>

vllm_ascend/attention/attention_v1.py

@@ -151,6 +151,7 @@ class AscendMetadata:
     slot_mapping: torch.Tensor = None
 
     enable_dbo_across_dp: bool = False
+    is_only_prefill: bool = False
 
 
 class AscendAttentionMetadataBuilder:
@@ -166,7 +167,8 @@ class AscendAttentionMetadataBuilder:
               num_reqs,
               num_actual_tokens,
               max_query_len,
-              enable_dbo_across_dp: bool = False):
+              enable_dbo_across_dp: bool = False,
+              is_only_prefill: bool = False):
 
         block_table = self.runner.input_batch.block_table[0].get_device_tensor(
         )
@@ -203,7 +205,8 @@ class AscendAttentionMetadataBuilder:
             slot_mapping=slot_mapping,
             attn_mask=attn_mask,
             attn_state=attn_state,
-            enable_dbo_across_dp=enable_dbo_across_dp)
+            enable_dbo_across_dp=enable_dbo_across_dp,
+            is_only_prefill=is_only_prefill)
         return attn_metadata
 
 

vllm_ascend/attention/attention_v1_torchair.py

@@ -223,7 +223,9 @@ class AscendAttentionTorchairMetadataBuilder:
               num_actual_tokens,
               max_query_len,
               graph_pad_size: int = -1,
-              enable_dbo_across_dp: bool = False):
+              enable_dbo_across_dp: bool = False,
+              *args,
+              **kwargs):
 
         device = self.runner.device
 

vllm_ascend/attention/mla_v1.py

@@ -384,6 +384,8 @@ class AscendMLAMetadataBuilder:
         graph_pad_size: int = -1,
         query_start_loc: torch.Tensor = None,
         enable_dbo_across_dp: bool = False,
+        *args,
+        **kwargs,
     ) -> AscendMLAMetadata:
         assert self._num_decodes + self._num_prefills == num_reqs