fix(ci): stop Flask server before uv cache cleanup

The Playwright CI was failing on push events because the background
Flask server (started via uv run flask) was never explicitly stopped.
The setup-uv action post-job cache pruning step would then hang for
5 minutes waiting on UV cache locks held by the orphan process,
failing with exit code 2.

Changes:
- Separate server startup into its own step with PID capture
- Add an always() cleanup step that stops the server before
  post-job steps run
- Keep the test step focused on running Playwright tests only

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: gkorland

.github/workflows/playwright.yml

@@ -81,17 +81,30 @@ jobs:
         npx playwright install chromium firefox
         npx playwright install-deps chromium firefox
 
-    - name: Run Playwright tests
+    - name: Start Flask server
+      id: start-server
       env:
         OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
         SECRET_TOKEN: ${{ secrets.SECRET_TOKEN }}
         CODE_GRAPH_PUBLIC: "1"
         MODEL_NAME: "openai/gpt-4.1-mini"
       run: |
         uv run flask --app api/index.py run --host 0.0.0.0 --port 5000 &
+        echo "pid=$!" >> "$GITHUB_OUTPUT"
         # Wait for Flask to be ready
         timeout 30 bash -c 'until curl -s http://localhost:5000/ > /dev/null 2>&1; do sleep 0.5; done'
-        npx playwright test --shard=${{ matrix.shard }}/2 --reporter=dot,list
+
+    - name: Run Playwright tests
+      env:
+        OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
+        SECRET_TOKEN: ${{ secrets.SECRET_TOKEN }}
+        CODE_GRAPH_PUBLIC: "1"
+        MODEL_NAME: "openai/gpt-4.1-mini"
+      run: npx playwright test --shard=${{ matrix.shard }}/2 --reporter=dot,list
+
+    - name: Stop Flask server
+      if: always()
+      run: kill ${{ steps.start-server.outputs.pid }} 2>/dev/null || true
 
     - name: Ensure required directories exist
       if: always()