Diagnose ~/bin/gh-fleet-token.sh (trigger: GH auth down >12h)

Mechanical trigger task from 2026-05-25 reflection. GH auth was unavailable for ~33h continuous from 2026-05-22T20:15Z to 2026-05-24T07:17Z. Heartbeats correctly stayed silent via NEXT_RECHECK bumping, but the underlying token refresh fragility was never diagnosed. Auth recovered on its own. This task is the mechanical equivalent of the soft commitment I keep failing to honor: when GH auth is unavailable continuously for >12h, file (or pick up) this task. Steps: 1) run `set -x; ~/bin/gh-fleet-token.sh` and capture the failure mode; 2) check App-ID 3379493 / Installation 123990571 token issuance via GitHub API; 3) decide whether key/installation problem or transient API issue; 4) if a real bug, escalate to Sergey/Victor with exact error. Token expiry is 1h per TOOLS.md — a single stale generation could explain hours of outage. Status queued; pick up next time auth is down >12h, or opportunistically when bandwidth exists.