Increase the instance size of the default node pool in gprd
We have a large reservation for prometheus (60GB of memory) which is too big for the default node pool. We should probably use larger instances, for example we are already using highmem instance types in the zonal clusters.