Runner Metrics tab now shows real per-job machine usage

Runner jobs now populate the Metrics tab with real CPU, memory, network, and disk usage. Sampling moved from the runners-controller into the runner VM itself (forked into the background on macOS, run as a dedicated sidecar on Linux) because the previous controller-side collector produced nothing, especially on the macOS Tart fleet whose custom kubelet couldn’t be observed from outside the VM. Samples are POSTed every 15 seconds for the full duration of the job and appear automatically once both the new runner images and the updated server authentication are deployed.