Long-running macOS runner jobs no longer truncate metrics mid-run

We increased the per-Pod runner service account token TTL from 1 hour to 8 hours. The macOS runner metrics sampler reuses this token for the whole job, but it was minted once at VM boot with a 1-hour TTL and never rotated. On warm Pods that sat idle before claiming a long job, the token expired mid-run, causing sample POSTs to return 401 and the metrics chart to stop partway through. The longer TTL keeps the token valid for the full duration of long jobs.