Compare performance between cascadelake and sapphirerapids queues after compiling with arch-specific flags

[manodeep]

Compile the models with compiler flags that contain the specific ISA (i.e., -mcpu=<name> (or, equivalently, -march=<name> -mtune=<name>) for name $\in$ [cascadelake, sapphirerapids]. Then run the models with the correct executable on the targeted queue, and compare performance. Likely requires updating payu to handle different queues (payu-org/payu#561)

Needs a bit of a think about how to make a fair comparison, since each node on the different queues likely will contain i) different cpus and/or ii) cost different SUs