[Platform][Refactor] Extract func get_default_attn_backend to Platform (#10358)

Signed-off-by: Mengqing Cao <cmq0113@163.com>

vllm/model_executor/models/utils.py

@@ -9,13 +9,13 @@ from torch.func import functional_call
 from transformers import PretrainedConfig
 
 import vllm.envs as envs
-from vllm.attention.selector import (_Backend, backend_name_to_enum,
+from vllm.attention.selector import (backend_name_to_enum,
                                      get_global_forced_attn_backend)
 from vllm.config import VllmConfig
 from vllm.logger import init_logger
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 from vllm.multimodal import MultiModalPlaceholderMap, NestedTensors
-from vllm.platforms import current_platform
+from vllm.platforms import _Backend, current_platform
 from vllm.sequence import IntermediateTensors
 from vllm.utils import is_pin_memory_available