Runner-job Metrics tab now shows in-VM machine metrics

Runner-job Metrics tabs now actually populate. Machine metrics (CPU, memory, network, disk) for macOS Tart and Linux kata runner jobs are now sampled from inside the runner VM for the full job duration and POSTed to the server, so the server-side Metrics dashboard shows real usage instead of staying empty. Sampling is best-effort and fail-open, so a bad sample or failed POST never blocks the job.