Fix Qwen3Next CI Config (#7561)

### What this PR does / why we need it?
This pr modifies qwen3Next nightly CI config. 
(1) Add a nightly CI .
(2) Set a more precise accuracy standard

- vLLM version: v0.18.0
- vLLM main:
6a9cceb219

Signed-off-by: Your Name <you@example.com>
Co-authored-by: Your Name <you@example.com>

.github/workflows/schedule_nightly_test_a3.yaml

@@ -282,6 +282,9 @@ jobs:
           - name: qwen3-30b-a3b-w8a8
             os: linux-aarch64-a3-4
             config_file_path: Qwen3-30B-A3B-W8A8.yaml
+          - name: qwen3-next-80b-a3b-instruct
+            os: linux-aarch64-a3-4
+            config_file_path: Qwen3-Next-80B-A3B-Instruct.yaml
           - name: qwen3-next-80b-a3b-instruct-w8a8
             os: linux-aarch64-a3-4
             config_file_path: Qwen3-Next-80B-A3B-Instruct-W8A8.yaml

tests/e2e/nightly/single_node/models/configs/Qwen3-Next-80B-A3B-Instruct-A2.yaml

@@ -24,6 +24,8 @@ _server_cmd: &server_cmd
   - "0.8"
   - "--max-num-seqs"
   - "64"
+  - "--compilation_config"
+  - '{"cudagraph_mode": "FULL_DECODE_ONLY"}'
 
 _benchmarks: &benchmarks
   perf:
@@ -44,8 +46,8 @@ _benchmarks: &benchmarks
     max_out_len: 32768
     batch_size: 32
     top_k: 20
-    baseline: 95
-    threshold: 5
+    baseline: 96
+    threshold: 3
 
 # ==========================================
 # ACTUAL TEST CASES

tests/e2e/nightly/single_node/models/configs/Qwen3-Next-80B-A3B-Instruct-W8A8.yaml

@@ -41,5 +41,5 @@ test_cases:
         dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_chat_prompt
         max_out_len: 32768
         batch_size: 32
-        baseline: 95
-        threshold: 5
+        baseline: 96
+        threshold: 3

tests/e2e/nightly/single_node/models/configs/Qwen3-Next-80B-A3B-Instruct.yaml

@@ -24,8 +24,8 @@ _server_cmd: &server_cmd
   - "0.8"
   - "--max-num-seqs"
   - "64"
-  - "--compilation-config"
-  - '{"cudagraph_capture_sizes": [64]}'
+  - "--compilation_config"
+  - '{"cudagraph_mode": "FULL_DECODE_ONLY"}'
 
 _benchmarks: &benchmarks
   perf:
@@ -46,8 +46,8 @@ _benchmarks: &benchmarks
     max_out_len: 32768
     batch_size: 64
     top_k: 20
-    baseline: 95
-    threshold: 5
+    baseline: 96
+    threshold: 3
 
 # ==========================================
 # ACTUAL TEST CASES