[Bugfix] Safeguard against missing backend in AttentionBackendEnum (#28846)

Signed-off-by: jesse <szxfml@gmail.com>
Signed-off-by: Song Zhixin <szxfml@gmail.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

vllm/attention/layer.py

@@ -310,7 +310,8 @@ class Attention(nn.Module, AttentionLayerBase):
             kv_sharing_target_layer_name,
             **extra_impl_args,
         )
-        self.backend = AttentionBackendEnum[self.attn_backend.get_name()]
+        backend_name = self.attn_backend.get_name()
+        self.backend = AttentionBackendEnum.__members__.get(backend_name)
         self.dtype = dtype
 
         # For cuda-alike (CUDA and ROCM) and cpu platforms, we control how