[E2E] Optimize nightly testcase. (#4886)

### What this PR does / why we need it?
Optimize nightly testcase.
Changes:
- tests/e2e/nightly/multi_node/config/models/Qwen3-235B-A3B.yaml: Add
accuracy and performance benchmark
- tests/e2e/models/configs/Qwen3-8B-Base.yaml: Delete
- tests/e2e/models/configs/internlm-7b.yaml: Change to
internlm3-8b-instruct
- tests/e2e/nightly/models/test_deepseek_r1_w8a8_eplb.py: Change to
DeepSeek-R1-0528-W8A8 model

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

Signed-off-by: menogrey <1299267905@qq.com>

tests/e2e/models/configs/internlm3-8b-instruct.yaml

@@ -0,0 +1,13 @@
+model_name: "Shanghai_AI_Laboratory/internlm3-8b-instruct"
+hardware: "Atlas A2 Series"
+tasks:
+- name: "ceval-valid"
+  metrics:
+  - name: "acc,none"
+    value: 0.42
+num_fewshot: 5
+max_model_len: 2048
+trust_remote_code: True
+dtype: "bfloat16"
+apply_chat_template: False
+fewshot_as_multiturn: False