[feature] fia support sliding windows (#5239)

Enable fia to support sliding window function and adapt to the Gemma3
model.

- vLLM version: release/v0.13.0
- vLLM main:
ad32e3e19c
---------
Signed-off-by: nsdie <yeyifan@huawei.com>

vllm_ascend/attention/utils.py

@@ -142,6 +142,8 @@ class AscendCommonAttentionMetadata(CommonAttentionMetadata):
 
     spec_attn_mask: torch.Tensor = None
 
+    swa_mask: torch.Tensor = None
+
     attn_state: Any = None
 
     graph_pad_size: int = -1
@@ -175,6 +177,7 @@ class AscendCommonAttentionMetadata(CommonAttentionMetadata):
             positions=self.positions[:num_actual_tokens],
             attn_mask=self.attn_mask,
             spec_attn_mask=self.spec_attn_mask,
+            swa_mask=self.swa_mask,
             attn_state=self.attn_state,
             graph_pad_size=-1,  # It should be -1 when not run in fullgraph mode.
             num_input_tokens=num_actual_tokens,