Long macOS runner jobs now show metrics for the whole job, not just the first portion

Long-running macOS runner jobs no longer have their machine metrics charts cut off partway through. Two underlying causes were addressed together: the per-Pod SA token the in-VM sampler reuses to POST samples was being minted at VM boot with a 1h TTL and never rotated, so a token aged out mid-job and samples stopped; the TTL is now 8h (still Pod-bound, so security isn’t reduced). Also, a newly booted macOS VM could report step timestamps off a clock that timed had not yet synced, by minutes at the start of the run, which made the step timeline and the metric timeline disagree; the VM now runs a best-effort NTP step right after dispatch so both clocks agree from the first sample. Affects macOS runners only.