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/ops/test_rotary_embedding.py

@@ -4,6 +4,7 @@
 # Adapted from
 # https://github.com/vllm-project/vllm/blob/main/vllm/tests/kernels/test_rotary_embedding.py
 
+import gc
 from typing import Optional, Tuple, Union
 
 import pytest
@@ -199,6 +200,9 @@ def test_rotary_embedding_quant_with_leading_dim(
                                ref_key,
                                atol=DEFAULT_ATOL,
                                rtol=DEFAULT_RTOL)
+    gc.collect()
+    torch.npu.empty_cache()
+    torch.npu.reset_peak_memory_stats()
 
 
 class ModelwithRotaryEmbedding(nn.Module):
@@ -342,3 +346,6 @@ def test_capture_rotary_embedding_in_aclgraph(
                                output_reference,
                                atol=DEFAULT_ATOL,
                                rtol=DEFAULT_RTOL)
+    gc.collect()
+    torch.npu.empty_cache()
+    torch.npu.reset_peak_memory_stats()