TEMPORARY: Test kind-deployment with max_connections = 500

Deploy Cloud Foundry from the ZPascal/kind-deployment fork, which has
cloudfoundry/kind-deployment#516, and drop the Postgres restart
workaround. Revert to the cloudfoundry/kind-deployment pin once cloudfoundry#516 is
merged.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

Author: ZPascal

.github/workflows/integration-test.yml

@@ -88,36 +88,14 @@ jobs:
           echo "file=$RUNNER_TEMP/capi-values.yaml" >> "$GITHUB_OUTPUT"
       # Every job gets its own Cloud Foundry on KinD, so the matrix can run in parallel.
       # The action deploys the kind-deployment ref given in `ref`, so keep both on the same commit.
+      # TEMPORARY: the fork contains cloudfoundry/kind-deployment#516 (max_connections = 500).
       - name: Set up Cloud Foundry
-        uses: cloudfoundry/kind-deployment/.github/actions/setup-cf@8eff1a34c730d89d3c90044695cbcaba048ec68d
+        uses: ZPascal/kind-deployment/.github/actions/setup-cf@6c4e9873608fe4bacf7815a9b15c075164d99d18
         with:
-          ref: 8eff1a34c730d89d3c90044695cbcaba048ec68d
+          ref: 6c4e9873608fe4bacf7815a9b15c075164d99d18
           use-latest-versions: ${{ (inputs.preview || github.event_name == 'schedule') }}
           github-token: ${{ github.token }}
           additional-values-files: ${{ steps.values.outputs.file }}
-      # Workaround: all Cloud Foundry components share one Postgres with the default max_connections
-      # (100). Over a full test run it runs out ("too many clients already") and UAA, which every
-      # Spring test context needs, stops working. kind-deployment doesn't expose the setting, and
-      # it only takes effect after a restart.
-      - name: Raise Postgres connection limit
-        timeout-minutes: 20
-        run: |
-          .github/scripts/pg-psql.sh -c "ALTER SYSTEM SET max_connections = 500"
-          kubectl -n cf-system rollout restart statefulset/postgresql
-          kubectl -n cf-system rollout status statefulset/postgresql --timeout=300s
-          .github/scripts/pg-psql.sh -tc "SHOW max_connections"
-          # The restart leaves the components with dead pooled connections (the cloud controller
-          # answers one 500 per stale connection and does not validate them), and some of them
-          # crash-loop for a while. Restart them all so they reconnect, wait for the rollouts and check
-          # that the platform answers before the tests start.
-          kubectl -n cf-system rollout restart deployment
-          for deployment in $(kubectl -n cf-system get deployment -o name); do
-            kubectl -n cf-system rollout status "$deployment" --timeout=600s
-          done
-          for url in "https://$CF_API_HOST/v3/info" "https://$CF_UAA_HOST/info"; do
-            curl --silent --show-error --fail --insecure --output /dev/null \
-              --retry 60 --retry-delay 10 --retry-all-errors "$url"
-          done
       - name: Set up Java
         uses: actions/setup-java@v5
         with: