The in-process token cache was lost on every app restart, forcing a new token
request each time and tripping Dell's token-endpoint rate limit. Add an L2 file
cache in the instance dir (keyed by a hash of the client id, so the id never
lands on disk), mirroring warranty_sync.py. Once one token is obtained it is
reused for ~1h across restarts and workers, avoiding the cooldown in normal use.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
b90a13c7e5