feat: auto-scale CPU nodes (2-30) instead of fixed 60

[wdvr]

Summary

• CPU ASGs (cpu-arm, cpu-x86) now scale dynamically between 2-30 nodes based on demand, instead of running 30 fixed instances each 24/7
• The availability updater Lambda manages scaling: scales up when spare slots < 2, scales down (with hysteresis) when spare slots > 8
• Instance protection prevents terminating nodes with active gpu-dev pods during scale-down
• CLI gpu-dev avail shows scalable capacity (6 / 90) and ~3min (scaling up) wait estimate

Files changed

File	Change
main.tf	CPU types: instance_count=2, min_instance_count=2, max_instance_count=30 (both workspaces)
eks.tf	ASG min_size/max_size use new fields via try(), lifecycle { ignore_changes = [desired_capacity] }
availability.tf	IAM: SetDesiredCapacity, SetInstanceProtection, ec2:DescribeInstances
availability_updater/index.py	Autoscaling logic, instance protection, scalable_total in DynamoDB
cli.py + reservations.py	Display scalable capacity and scale-up time estimate

Test plan

• terraform plan — confirm CPU ASGs change from 30/30/30 to 2/2/30, GPU ASGs unchanged
• After tf apply: CPU nodes scale down to 2 each
• Scale-up test: reserve CPU slot → Lambda detects < 2 spare → sets desired +1 → node joins in ~3min
• Scale-down test: cancel all CPU reservations → Lambda detects > 8 spare → protects occupied nodes → reduces desired
• gpu-dev avail shows total / scalable_total format for CPU types and ~3min (scaling up) when at 0
• GPU ASGs remain completely unaffected (no min/max_instance_count fields = try() falls back to instance_count)