[Model] Support pooling models (#3122)

### What this PR does / why we need it?

Support pooling models (like `bge-reranker-v2-m3`) in vllm-ascend, this
pr covered the three model types of embed (cls_token, mean_token,
lasttoken).

After this
[commit](17373dcd93),
vllm has provided support for adapting pooling models on the v1 engine.
This PR includes corresponding adaptations on the vllm-ascend side.

Fixes #1960

- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c

---------

Signed-off-by: lianyibo <lianyibo1@kunlunit.com>
Signed-off-by: MengqingCao <cmq0113@163.com>
Co-authored-by: MengqingCao <cmq0113@163.com>

tests/e2e/singlecard/pooling/test_classification.py

@@ -0,0 +1,34 @@
+import torch
+from modelscope import snapshot_download  # type: ignore[import-untyped]
+from transformers import AutoModelForSequenceClassification
+
+from tests.e2e.conftest import HfRunner, VllmRunner
+
+
+def test_classify_correctness() -> None:
+
+    model_name = snapshot_download("Howeee/Qwen2.5-1.5B-apeach")
+
+    prompts = [
+        "Hello, my name is",
+        "The president of the United States is",
+        "The capital of France is",
+        "The future of AI is what",
+    ]
+    with VllmRunner(
+            model_name,
+            runner="pooling",
+            max_model_len=None,
+            cudagraph_capture_sizes=[4],
+    ) as vllm_runner:
+        vllm_outputs = vllm_runner.classify(prompts)
+
+    with HfRunner(model_name,
+                  dtype="float32",
+                  auto_cls=AutoModelForSequenceClassification) as hf_runner:
+        hf_outputs = hf_runner.classify(prompts)
+
+    for hf_output, vllm_output in zip(hf_outputs, vllm_outputs):
+        hf_output = torch.tensor(hf_output)
+        vllm_output = torch.tensor(vllm_output)
+        assert torch.allclose(hf_output, vllm_output, 1e-2)