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_profile_execute_duration.py

@@ -16,6 +16,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 #
+import gc
 import os
 import time
 from unittest.mock import patch
@@ -50,6 +51,10 @@ def test_execue_duration_enabled_discrepancy():
     assert diff <= 0.5, (
         f"CPU={cpu_duration:.2f}ms, NPU={npu_durations['forward']:.2f}ms")
 
+    gc.collect()
+    torch.npu.empty_cache()
+    torch.npu.reset_peak_memory_stats()
+
 
 def test_execue_duration_disabled():
     a = torch.randn(100, 100).npu()
@@ -60,3 +65,7 @@ def test_execue_duration_disabled():
         torch.npu.synchronize()
     npu_durations = ProfileExecuteDuration().pop_captured_sync()
     assert not npu_durations
+
+    gc.collect()
+    torch.npu.empty_cache()
+    torch.npu.reset_peak_memory_stats()