Add GitHub workflow for integration tests on multiple Java versions

[ZPascal]

This adds a workflow that runs the integration tests against a Cloud Foundry on KinD, using kind-deployment, on Java 8, 11, 17 and 21. Each matrix job gets its own Cloud Foundry, so they can run in parallel.

The workflow is integration-test.yml. It runs ./mvnw -Pintegration-test test against the deployed CF and is triggered

• nightly on weekdays (latest versions, Java 21 only, and only in cloudfoundry/cf-java-client, not on forks),
• manually via workflow_dispatch, optionally with preview,
• from ci-java.yml after the build job (Java 21).

Normal runs deploy the capi-release that matches SUPPORTED_API_VERSION. .github/scripts/resolve-capi-version.sh finds it and the workflow hands it to kind-deployment as a chart version override. Preview runs (nightly, or manual with preview) just take the latest versions.

PostgreSQL connections

All CF components share a single PostgreSQL, and its default of 100 connections runs out during a full test run (too many clients already). Once that happens UAA fails and so does every test context after it. The workflow therefore sets postgres-max-connections: '500', which needs cloudfoundry/kind-deployment#516.

Until #516 is merged, setup-cf and ref point to a commit on my fork (ZPascal/kind-deployment) that contains it. Both are marked TEMPORARY in the workflow and need to be moved to the upstream commit afterwards.

Dependencies

• feat: Make the max_connections of the shared PostgreSQL configurable kind-deployment#516, see above.
• The integration tests need a few client and test changes to work with a current CAPI, for example the resources field on bits uploads and the TCP port range of the router group. Those are not part of this PR. For 5.x.x they are in Adapt the client and the integration tests to the KinD based CAPI #1383. Until they are merged, the integration job in this PR is expected to fail.
• The workflow sets TEST_SKIP_METRIC_REGISTRAR and TEST_QUOTAS_ROUTES_RESERVEDPORTS. The names have to match what the integration tests on the target branch read. On 5.x.x those are SKIP_METRIC_REGISTRAR_TESTS and TEST_QUOTAS_ROUTES_RESERVED_PORTS, so the workflow needs adjusting on one side or the other.

[Lokowandtg]

The failures because of "suspended" where addressed with #1371

[Lokowandtg]

The failures because of "missing :resources" might be a problem in kind? As a workaround I added the following line to https://github.com/cloudfoundry/cf-k8s-releases/blob/4500c9dec5007c5c86c630bb86c3bb596e98259c/capi/helm/files/nginx/nginx_external_endpoints.conf#L47
upload_set_form_field "resources" [];
resulting in running tests.