[refactor] refactor model runner capture model (#5230)

### What this PR does / why we need it?
Refactor the `capture_model` method in model_runner to directly reuse
the method from vLLM.

Currently, most of the logic in the capture_model method is similar to
that in the vllm code. Directly using the vllm method can reduce the
maintenance cost of the vllm-ascend code. Modify as follows:
1、refactor capture_model function, directly inheriting community methods
2、refactor initialize_aclgraph_capture function, move to
initialize_attn_backend

### Does this PR introduce _any_ user-facing change?
No

### How was this patch tested?

- vLLM version: release/v0.13.0
- vLLM main:
ad32e3e19c

Signed-off-by: weiguihua2 <weiguihua2@huawei.com>
Co-authored-by: weijinqian0 <1184188277@qq.com>

vllm_ascend/attention/mla_v1.py

@@ -1,6 +1,5 @@
 from dataclasses import dataclass
-from typing import (TYPE_CHECKING, ClassVar, NamedTuple, Optional, Tuple, Type,
-                    TypeVar)
+from typing import TYPE_CHECKING, NamedTuple, Optional, Tuple, Type, TypeVar
 
 import numpy as np
 import torch
@@ -15,7 +14,7 @@ from vllm.model_executor.layers.linear import UnquantizedLinearMethod
 from vllm.utils.math_utils import cdiv, round_down
 from vllm.v1.attention.backends.mla.common import MLACommonMetadataBuilder
 from vllm.v1.attention.backends.utils import AttentionCGSupport
-from vllm.v1.kv_cache_interface import MLAAttentionSpec
+from vllm.v1.kv_cache_interface import AttentionSpec, MLAAttentionSpec
 
 from vllm_ascend import envs
 from vllm_ascend.ascend_config import get_ascend_config
@@ -182,9 +181,6 @@ M = TypeVar("M", bound=AscendMLAMetadata)
 
 
 class AscendMLAMetadataBuilder(MLACommonMetadataBuilder[AscendMLAMetadata]):
-    # Does this backend/builder support ACL Graphs for attention (default: no).
-    aclgraph_support: ClassVar[AttentionCGSupport] = \
-        AttentionCGSupport.UNIFORM_BATCH
     """
     NOTE: Please read the comment at the top of the file before trying to
     understand this class
@@ -263,6 +259,16 @@ class AscendMLAMetadataBuilder(MLACommonMetadataBuilder[AscendMLAMetadata]):
         self.query_lens: torch.Tensor = None
         self.seq_lens: torch.Tensor = None
 
+    @classmethod
+    def get_cudagraph_support(
+        cls: type["AscendMLAMetadataBuilder"],
+        vllm_config: VllmConfig,
+        kv_cache_spec: AttentionSpec,
+    ) -> AttentionCGSupport:
+        # Explicit override in case the underlying builder specialized this getter.
+        # @override omitted only because of mypy limitation due to type variable.
+        return AttentionCGSupport.UNIFORM_BATCH
+
     def reorder_batch(self, input_batch: "NPUInputBatch",
                       scheduler_output: "SchedulerOutput") -> bool:
         # We now want to reorder the batch so that the "decode" requests are at