Simplify performance regression test workflow

- Remove complex pytest logic that was causing CI failures
- Run performance regression test directly using the evaluator script
- Add proper error handling and verification of test output
- Ensure CI fails appropriately if performance thresholds are exceeded

This simplifies the workflow and makes it more reliable by directly
testing the evaluator functionality rather than relying on pytest markers.

Author: khaosans

.github/workflows/verify.yml

@@ -225,20 +225,16 @@ jobs:
           OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
           OPENAI_MODEL: ${{ vars.OPENAI_MODEL || 'gpt-3.5-turbo' }}
         run: |
-          # Parallelize for speed
-          TEST_OUTPUT=$(poetry run pytest -n auto tests/ -m "performance" -v --tb=short 2>&1)
-          TEST_EXIT_CODE=$?
-          echo "$TEST_OUTPUT"
+          # Run performance regression test directly
+          echo "Running performance regression test..."
+          poetry run python scripts/test_performance_regression.py
           
-          # Check if any tests were actually run
-          if echo "$TEST_OUTPUT" | grep -q "no tests ran"; then
-            echo "No performance tests found, running fallback script..."
-            poetry run python scripts/test_performance_regression.py
-          elif [ $TEST_EXIT_CODE -ne 0 ]; then
-            echo "Performance tests failed, running fallback script..."
-            poetry run python scripts/test_performance_regression.py
+          # Verify the test output
+          if [ $? -eq 0 ]; then
+            echo "✅ Performance regression test completed successfully"
           else
-            echo "Performance tests completed successfully"
+            echo "❌ Performance regression test failed"
+            exit 1
           fi
       - name: Upload Performance Metrics
         if: always()