Refactor e2e CI (#2276)

Refactor E2E CI to make it clear and faster
1. remove some uesless e2e test
2. remove some uesless function
3. Make sure all test runs with VLLMRunner to avoid oom error
4. Make sure all ops test end with torch.empty_cache to avoid oom error
5. run the test one by one to avoid resource limit error


- vLLM version: v0.10.1.1
- vLLM main:
a344a5aa0a

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>

tests/e2e/singlecard/test_guided_decoding.py

@@ -28,7 +28,7 @@ from vllm.sampling_params import GuidedDecodingParams, SamplingParams
 from tests.e2e.conftest import VllmRunner
 
 os.environ["PYTORCH_NPU_ALLOC_CONF"] = "max_split_size_mb:256"
-MODEL_NAME = "Qwen/Qwen2.5-0.5B-Instruct"
+MODEL_NAME = "Qwen/Qwen3-0.6B"
 
 GuidedDecodingBackend = ["xgrammar", "guidance", "outlines"]
 
@@ -92,7 +92,6 @@ def test_guided_json_completion(guided_decoding_backend: str,
     with VllmRunner(
             MODEL_NAME,
             seed=0,
-            dtype="auto",
             guided_decoding_backend=guided_decoding_backend,
     ) as vllm_model:
         prompts = [
@@ -131,7 +130,6 @@ def test_guided_regex(guided_decoding_backend: str, sample_regex):
     with VllmRunner(
             MODEL_NAME,
             seed=0,
-            dtype="auto",
             guided_decoding_backend=guided_decoding_backend,
     ) as vllm_model:
         prompts = [