feat: enable in-place pod vertical scaling

[mpeterson]

Background

Kubernetes 1.35 graduated in-place pod vertical scaling to GA. You can now modify CPU and memory requests/limits for running pods without restarting them.

Current State

• Resource limits are static — changing them requires pod restart
• Workloads like Ollama and Plex have variable resource needs

Opportunity

• Ollama: Scale GPU memory/CPU dynamically based on model loading
• Plex: Increase resources during heavy transcoding, reduce when idle
• VPA integration: Vertical Pod Autoscaler can now resize without restarts
• Reduces disruption for stateful workloads using iSCSI PVCs

Prerequisites

• Cluster running K8s ≥ 1.35
• InPlacePodVerticalScaling feature gate (GA in 1.35, enabled by default)

Action

Identify workloads that would benefit most from dynamic resizing. Consider deploying VPA with in-place update mode.