[CPU] Fix macOS/Apple Silicon hang by enabling OpenMP in the build (#46769)

Signed-off-by: mgoin <mgoin64@gmail.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

.github/actionlint.yaml

@@ -3,3 +3,5 @@
 self-hosted-runner:
   labels:
     - vllm-runners
+    # Not yet in actionlint's known-label set.
+    - macos-26

.github/workflows/macos-smoke-test.yml

@@ -11,7 +11,19 @@ permissions:
 
 jobs:
   macos-m1-smoke-test:
-    runs-on: macos-latest
+    # macos-26 (the supported target) is still a preview runner, so gate on GA
+    # macos-15 and keep macos-26 non-blocking.
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - os: macos-15
+            required: true
+          - os: macos-26
+            required: false
+    name: macos-m1-smoke-test (${{ matrix.os }})
+    runs-on: ${{ matrix.os }}
+    continue-on-error: ${{ !matrix.required }}
     timeout-minutes: 30
 
     steps:
@@ -72,14 +84,11 @@ jobs:
           # Test health endpoint
           curl -f http://localhost:8000/health
 
-          # Test completion
-          curl -f http://localhost:8000/v1/completions \
+          # Long prompt: hits the split-KV path that short prompts skip (#46769).
+          PAYLOAD=$(python -c "import json; print(json.dumps({'model': 'Qwen/Qwen3-0.6B', 'prompt': 'The quick brown fox jumps over the lazy dog. ' * 24, 'max_tokens': 16}))")
+          curl -f --max-time 120 http://localhost:8000/v1/completions \
             -H "Content-Type: application/json" \
-            -d '{
-              "model": "Qwen/Qwen3-0.6B",
-              "prompt": "Hello",
-              "max_tokens": 5
-            }'
+            -d "$PAYLOAD"
 
           # Cleanup
           kill "$SERVER_PID"