Reduce cost for pangeo-hubs, three n2-highmem-8 core nodes running

[consideRatio]

The pods running could fit on a single node computational capacity wise, I think they don't run on a single node because of a few pods with anti-affinity policies for example. This can be improved on greatly no matter what.

I think this must be related to having many konnectivity-agents that doesn't want to be scheduled together on nodes more than "1 skew", discussed in #2490. I reduced the autoscaling of these pods to not have more than 2 until at least the cluster had 10 nodes running to help avoid the need for more than 1 core node in most situations.

[consideRatio]

I've managed to get this down from 3 to 1 nodes, and it can be further reduced to a single n2-highmem-4 node instead of using the n2-highmem-8 nodes. A comment that this is more suitable is added in #3375.

[consideRatio]

Remaining action point

• Change the core nodes for pangeo-hubs from n2-highmem-8 nodes to n2-highmem-4 nodes

[consideRatio]

• Fixed by pangeo-hubs: upgrade k8s cluster, use smaller core nodes #3409