I would caution you here. First of all you want to do extreme vetting on any new gateway version as they tend to come out half baked, and there is no clean way to downgrade (only a complete remove and reinstall of the LKG version will do).
For auto scaling you need to have VMs already prepared including the gateway service, the configuration changes for the proxy, the ADO and ODBC drivers - everything MUST match between cluster members, at all times.
You might get away with a cloned VM image for that but the machine name itself would have to be hand configured.
The actual onboarding and offboarding of cluster members in the cluster is not something you can do programmatically. You need to specify the target gateway, inject the Recovery key, and confirm the action - all manually, on each of the cluster members.
Once you have a pool of cluster members you can use the web interface to selectively disable some of them. However again this is a manual process without any automation options.
What you will want is extensive monitoring and telemetry to gauge if you actually need to increase the number of cluster members, or if updating the memory or CPU count on the existing VMs is sufficient. Check the average mashup container count in your telemetry.
