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/multicard/test_qwen3_moe.py

@@ -32,11 +32,9 @@ def test_models_distributed_Qwen3_MOE_TP2():
     example_prompts = [
         "Hello, my name is",
     ]
-    dtype = "half"
     max_tokens = 5
     with VllmRunner(
             "Qwen/Qwen3-30B-A3B",
-            dtype=dtype,
             tensor_parallel_size=2,
             distributed_executor_backend="mp",
     ) as vllm_model:
@@ -47,11 +45,9 @@ def test_models_distributed_Qwen3_MOE_TP2_WITH_EP():
     example_prompts = [
         "Hello, my name is",
     ]
-    dtype = "half"
     max_tokens = 5
     with VllmRunner(
             "Qwen/Qwen3-30B-A3B",
-            dtype=dtype,
             tensor_parallel_size=2,
             enable_expert_parallel=True,
             distributed_executor_backend="mp",
@@ -64,12 +60,10 @@ def test_models_distributed_Qwen3_MOE_W8A8():
     example_prompts = [
         "Hello, my name is",
     ]
-    dtype = "auto"
     max_tokens = 5
     with VllmRunner(
             snapshot_download("vllm-ascend/Qwen3-30B-A3B-W8A8"),
             max_model_len=8192,
-            dtype=dtype,
             tensor_parallel_size=2,
             quantization="ascend",
             enforce_eager=True,