Main reason is that PR won't have the secret to post to the webhook URL.
In addition, there's someone looking at PRs, at least the author, and
therefore the failure will be noticed.

This allows us to test CI changes in a dedicated branch, and keep the
main branch history clean.

This allows to run them in parallel, unblocking other jobs that depend
on the build_operator job.

Unfortunately, we cannot get rid of this step, because we need to
install testenv, and the setup-env binary for testing.

We are pushing every commit to a registry. This seems a bit unnecessary,
specially given that we can build locally very quickly. It's more
efficient to build and export to a tarball, and then share this tarball
among other jobs.

The GCS and manifest part is now a separate job, so it doesn't delay the
execution of other parts of the workflow. kubectl-rabbitmq plugin tests
now use the locally built image, instead of locally building one. This
should speed-up the execution time of this job.

The examples test.sh script was not exiting when an error was returned.
Any error in the test.sh scripts should now stop the execution and
return the exit code.

__ Downscale vault-tls example
The github runners struggles to run 3-node rabbit cluster due to
resource constraints.

__ Delete obsolete example
Federation is better achieved using the Messaging Topology Operator.
There's another example that covers how to setup TLS. The case of
federation over TLS is a combination of using the Topology Operator in
combination with the TLS example.

__ Fix external secret example
The test.sh file was missing, and this test was not excluded from the CI
runs. Added a short script to validate that the external secret is used
to seed the default user.

__ Removed system tests in GKE
They do not provide more value than local KinD system tests

__ cmctl now waits watching the correct namespace
Earlier, if the kubeconfig had a namespace that didn't exist as current
ns, cmctl will always fail, even when cert-manager was ready

__ Skip mtls internode example tests

It requires a 3 node cluster, and this setup is not reliable inside a
GitHub runner, given the resource constraints inside the runner.

Typo in kubectl testss

Starting with cert-manager 1.15.x, cmctl cli and cert manager are
shipped in different repos, as different software.

To avoid rate limitting from GitHub

This allows the setup-go action to cache our dependencies. In order to
cache our deps, the go.sum file needs to be present before executing the
setup-go action.

Login to Dockerhub, otherwise, the push will fail.

[skip ci]