[Feat][Graph] Support MTP for ACL Graph (#2932)

### What this PR does / why we need it?
This PR depends on the merge of #2707 and has adapted the aclgraph
functionality to support MTP.

### How was this patch tested?


- vLLM version: v0.10.2
- vLLM main:
2b85697031

---------

Signed-off-by: xuyexiong <xuyexiong@huawei.com>

tests/e2e/singlecard/spec_decode_v1/test_v1_mtp_correctness.py

@@ -39,7 +39,7 @@ def mtp_correctness(
                     tensor_parallel_size=1,
                     gpu_memory_utilization=0.7,
                     max_model_len=256,
-                    enforce_eager=True) as ref_llm:
+                    enforce_eager=False) as ref_llm:
         ref_outputs = ref_llm.generate(example_prompts, sampling_config)
 
     with VllmRunner(
@@ -53,7 +53,7 @@ def mtp_correctness(
                 "method": "deepseek_mtp",
                 "num_speculative_tokens": num_speculative_tokens,
             },
-            enforce_eager=True,
+            enforce_eager=False,
             max_model_len=2000,
             additional_config={"ascend_scheduler_config": {
                 "enabled": False

tests/ut/attention/test_mla_v1.py

@@ -186,6 +186,34 @@ class TestAscendMLAMetadataBuilder(TestBase):
         mock_vllm_config.scheduler_config.chunked_prefill_enabled = False
         mock_device = 'cpu'
 
+        mock_vllm_config.speculative_config = None
+
+        ascend_config = MagicMock()
+        with patch("vllm_ascend.attention.mla_v1.get_ascend_config",
+                   return_value=ascend_config):
+            builder = AscendMLAMetadataBuilder(None, None, mock_vllm_config,
+                                               mock_device)
+
+            self.assertEqual(builder.block_size,
+                             mock_vllm_config.cache_config.block_size)
+            self.assertEqual(
+                builder.chunked_prefill_enabled,
+                mock_vllm_config.scheduler_config.chunked_prefill_enabled)
+
+    def test_ascend_mla_metadata_builder_spec_decode(self):
+        mock_vllm_config = MagicMock()
+        mock_vllm_config.model_config.max_model_len = 1024
+        mock_vllm_config.model_config.get_head_size.return_value = 64
+        mock_vllm_config.model_config.dtype = torch.float16
+        mock_vllm_config.cache_config.block_size = 16
+        mock_vllm_config.scheduler_config.max_num_seqs = 4
+        mock_vllm_config.scheduler_config.chunked_prefill_enabled = False
+        mock_device = 'cpu'
+
+        mock_spec_config = MagicMock()
+        mock_spec_config.num_speculative_tokens = 3
+        mock_vllm_config.speculative_config = mock_spec_config
+
         ascend_config = MagicMock()
         with patch("vllm_ascend.attention.mla_v1.get_ascend_config",
                    return_value=ascend_config):
@@ -208,6 +236,8 @@ class TestAscendMLAMetadataBuilder(TestBase):
         mock_vllm_config.scheduler_config.chunked_prefill_enabled = False
         mock_device = 'cpu'
 
+        mock_vllm_config.speculative_config = None
+
         with patch("vllm_ascend.attention.mla_v1.get_ascend_config",
                    return_value=ascend_config):
             builder = AscendMLAMetadataBuilder(None, None, mock_vllm_config,

tests/ut/torchair/test_torchair_mla.py

@@ -190,6 +190,8 @@ class TestAscendMLATorchairMetadataBuilder(TestBase):
         mock_vllm_config.scheduler_config.chunked_prefill_enabled = False
         mock_device = 'cpu'
 
+        mock_vllm_config.speculative_config = None
+
         ascend_config = MagicMock()
         ascend_config.torchair_graph_config = MagicMock()
         ascend_config.torchair_graph_config.enabled = True
@@ -217,6 +219,8 @@ class TestAscendMLATorchairMetadataBuilder(TestBase):
         ascend_config.torchair_graph_config = MagicMock()
         ascend_config.torchair_graph_config.enabled = True
 
+        mock_vllm_config.speculative_config = None
+
         builder = AscendMLATorchairMetadataBuilder(None, None,
                                                    mock_vllm_config,
                                                    mock_device)
@@ -252,6 +256,8 @@ class TestAscendMLATorchairMetadataBuilder(TestBase):
         mock_vllm_config.scheduler_config.chunked_prefill_enabled = False
         mock_device = 'cpu'
 
+        mock_vllm_config.speculative_config = None
+
         with patch("vllm_ascend.torchair.torchair_mla.get_ascend_config",
                    return_value=ascend_config):
             builder = AscendMLATorchairMetadataBuilder(None, None,
@@ -288,6 +294,8 @@ class TestAscendMLATorchairMetadataBuilder(TestBase):
         mock_vllm_config.scheduler_config.chunked_prefill_enabled = False
         mock_device = 'cpu'
 
+        mock_vllm_config.speculative_config = None
+
         builder = AscendMLATorchairMetadataBuilder(None, None,
                                                    mock_vllm_config,
                                                    mock_device)
@@ -309,6 +317,8 @@ class TestAscendMLATorchairMetadataBuilder(TestBase):
         mock_vllm_config.scheduler_config.chunked_prefill_enabled = False
         mock_device = 'cpu'
 
+        mock_vllm_config.speculative_config = None
+
         builder = AscendMLATorchairMetadataBuilder(None, None,
                                                    mock_vllm_config,
                                                    mock_device)
@@ -331,6 +341,8 @@ class TestAscendMLATorchairMetadataBuilder(TestBase):
         mock_vllm_config.scheduler_config.chunked_prefill_enabled = False
         mock_device = 'cpu'
 
+        mock_vllm_config.speculative_config = None
+
         builder = AscendMLATorchairMetadataBuilder(None, None,
                                                    mock_vllm_config,
                                                    mock_device)
@@ -357,6 +369,8 @@ class TestAscendMLATorchairMetadataBuilder(TestBase):
         mock_vllm_config.model_config.dtype = torch.float16
         mock_device = 'cpu'
 
+        mock_vllm_config.speculative_config = None
+
         builder = AscendMLATorchairMetadataBuilder(
             None,
             None,
@@ -424,6 +438,8 @@ class TestAscendMLATorchairMetadataBuilder(TestBase):
         model = MagicMock(spec=nn.Module)
         model.model = MagicMock(spec=nn.Module)
 
+        mock_vllm_config.speculative_config = None
+
         builder = AscendMLATorchairMetadataBuilder(
             None,
             None,