[Dashboard] Add memory graphs optimized for OOM debugging

[alanwguo]

Description

The current graph shows memory usage of each node along side the MAX memory across the cluster.

For OOM detection, we probably care more about % memory usage per node and should put extra emphasis on nodes with > 80% (or some other number) memory usage.

Use case

Debug OOM as reason my workload crashed

[alanwguo]

Another thing we can try doing is make head node emphasized in the Grafana dashboards. Make the line double thick.

[Bye-legumes]

Let me try to solve this, I prefer to head node emphasized with double thick.