Move some saturation recording rules to Thanos

Those saturation rules combine metrics from the deployment's Service monitor and the kubelet service monitor, which doesn't work with Prometheus shards (because they don't scrape kubelet metrics, which would result in duplicate metrics), so we need to evaluate them in Thanos instead.

See https://gitlab.com/gitlab-com/gl-infra/reliability/-/issues/16069

Unblocks gitlab-com/gl-infra/k8s-workloads/gitlab-helmfiles!861 (merged)