Runner-job Metrics tab now populates with in-VM metrics

The Metrics tab on runner jobs now actually shows machine metrics (CPU, memory, network, disk) for the duration of each job. Previously the tab stayed empty in production because the controller-side collector could not observe runner VM usage, especially on the macOS Tart fleet where nothing outside the VM can read cAdvisor data. Sampling has been moved into the runner itself: macOS Tart VMs run a background sampler that reports whole-VM CPU, memory, network and Data-volume disk usage every 15 seconds, and Linux runner microVMs run a dedicated metrics sidecar that reads VM-wide /proc for CPU/memory/network and the JIT volume for disk. Samples are best-effort and never affect job execution.