Add hyperdisk-ml support for gke-storage

[saltysoup]

Hyperdisk-ml is the recommended storage for speeding up load times and inference of larger models eg. DeepSeek-R1, Llama3-405b, which often takes a long time ~20min+

Source: https://cloud.google.com/kubernetes-engine/docs/tutorials/serve-multihost-gpu#speed_up_model_load_times_with_hyperdisk_ml

This PR is to add hyperdisk-ml support by adding this disk type as a valid condition in variables.tf, and new templates for storage class and PVC

[wiktorn]

Was this for really meant to be that long?

[saltysoup]

what do you mean by long? FWIW i used the same template as hyperdisk-extreme (noticed its slightly different to hyperdisk-balanced template).

Hyperdisk-ml doesnt support customizable IOPS so i took that out. I also didnt include custom throughput as it automatically specifies this based on disk size

[wiktorn]

I expected something like this:

Suggested change

	%{~ for key, val in labels ~}
	${key}: ${val}
	provisioner: pd.csi.storage.gke.io
	allowVolumeExpansion: true
	parameters:
	%{~ endfor ~}
	%{~ for key, val in labels ~}
	${key}: ${val}
	%{~ endfor ~}
	provisioner: pd.csi.storage.gke.io
	allowVolumeExpansion: true
	parameters:

I think you don't need to repeat provsioner and allowVolmeExpansion, and most imporatantly, you don't want them missing if no labels are provided. And to me it looks like hyperdisk-extreme has the same issue