Document limitations of the kubernetes node editing process

When performing this change: https://ops.gitlab.net/gitlab-com/gitlab-com-infrastructure/-/merge_requests/1902 we noticed that TF needed to recreate a node pool. It did not do it by adding a new nodepool, migrating workloads and then shutting down the old node pool. Instead, it evicted all workloads from the existing node pool, shut down all machines, removed the node pool (at which point there were lots of unscheduled pods) and then created a new node pool, added it to the cluster and scheduled workloads that were pending.

This is an example of where manually rolling over node pools would be preferred over using TF.

This is also a different scenario from a case where a TF change does not require a node pool recreation, but just editing (as was the case when enabling Workload Identity). In that scenario, nodes are added, workloads migrated across and old nodes are removed.

To be determined if this is a limitation on the GKE side or on the TF provider side.

In any case, this behaviour should be at least documented.