[backups] holds stale path; scheduled Kopia backups fail until external drive remount

**Summary:** On umbrelOS 1.5 with backups pointed at an external USB SSD, [backups] gets stuck: every hourly run fails with cannot access storage path ... no such file or directory even though the drive is mounted and the path is fully readable from a shell. Manual UI backup fails with the same error. Only a fresh mount event (e.g. unplug/replug) triggers Forgetting repository -> Connected to repository and restores normal operation. **Evidence that the filesystem is fine:** While umbreld reports ENOENT on the backup path: ``` $ ls "/home/umbrel/umbrel/external/Extreme SSD (2)/Umbrel.2/Umbrel Backup.backup"kopia.repository.f  kopia.maintenance.f  p/  q/  s/  x/ $ mount | grep external/dev/sdb1 on ... type exfat (rw,...,errors=remount-ro) $ df -h "/home/umbrel/umbrel/external/Extreme SSD (2)"/dev/sdb1  3.7T  623G  3.1T  17% ``` So the kopia child process spawned by umbreld sees a different view than a plain shell: stale path, namespace, or cached handle outliving a prior mount event. Also, the two modules use different path prefixes for the same filesystem (/External/... vs /home/umbrel/umbrel/external/...), which is where stale state plausibly accumulates. **Pattern:** - 9 consecutive hourly intervals failed at :31, identical error each time. Drive was never disconnected during that window. - Recovery only after USB re-enumeration -> [files:externalstorage] remount -> [backups] Forget/Connect. - Reproducible; has happened multiple times. **Expected:** When [files:externalstorage] (re)mounts the backup target, [backups] should invalidate cached state and reconnect automatically. Consecutive failures against the same repo should self-heal (Forget -> Connect) rather than retry a stuck handle indefinitely. **Environment:** Umbrel Home (2025), umbrelOS 1.5, SanDisk Extreme 4TB SSD (exFAT, USB 3.2 Gen 2).