Runner job metrics are now populated from in-VM sampling

The runner-job Metrics tab now shows live machine metrics by sampling CPU, memory, network, and disk usage from inside the runner VM itself. Previously the metrics collector ran in the runners-controller, but it could not read resource usage from the macOS Tart fleet or Linux kata nodes, so every job’s Metrics tab stayed empty. Metrics now post directly from the runner VM, so you can see actual resource utilization for the jobs you run.