Support autoscaling (HPA) from datumctl compute #293

Today datumctl compute deploy and datumctl compute scale only let you set a fixed minimum number of instances (--min). If you want workloads to automatically scale up and down based on CPU or memory usage, there's no CLI way to do it — you'd have to hand-edit the Workload object directly. This isn't discoverable and isn't something we expect users to do.

The underlying autoscaling support already exists in the platform - we just don't expose it through the CLI yet.

Proposal

Add autoscaling flags to deploy and scale:

• --max — maximum number of instances to scale up to
• --cpu-percent — target average CPU utilization to scale on
• --memory-percent — target average memory utilization to scale on

Example:

datumctl compute deploy api --min=2 --max=10 --cpu-percent=70
datumctl compute scale api --max=10 --cpu-percent=70

Setting --max and at least one of --cpu-percent/--memory-percent turns autoscaling on for the workload. Setting --max=0 turns it back off, reverting to a fixed instance count.

Why this matters

Autoscaling is a baseline expectation for anyone running production workloads, and doing it manually via scale calls is not a real substitute.

Open questions

• Should --min still be required on scale, or should adjusting just --max/--cpu-percent be allowed on its own?
• Do we want per-region/per-city autoscaling targets in this first pass, or one setting applied everywhere (matching how --min works today)?