Use cached version of the chart artifact between pipeline runs

As a followup to gitlab-com/gl-infra/k8s-workloads/gitlab-com!633 (merged) we should look to using a cached version of the built chart artifact between multiple pipeline runs. This will short-circuit the build step in the event that charts.gitlab.io is unavailable.

The was the scenario we saw during production#3875 (closed) which prevented us from deploying kubernetes.

Here is an example pipeline of the failure we saw (which is not surprising) when charts.gitlab.io was unavailable https://ops.gitlab.net/gitlab-com/gl-infra/k8s-workloads/gitlab-com/-/pipelines/506061