fix: state-integrity pass on the update loop (detail by name, honest checks, badge lifecycle)

[MotherSphere]

Why

Five confirmed defects where the UI misreports state or loses it.

Changes

1. The detail page is tracked by name, not index. The catalog vector is replaced AND reordered by every refresh (GitHub sorts by last push), so a refresh while a detail page was open silently swapped it to another app. Selection now goes by name through App::active_repo(), and a repository that disappeared falls back to the grid.
2. Update badges are cleared. After a successful update the card showed "Update vX → vX" until the next full check, and an uninstall kept its badge too.
3. An honest launcher check. check_launcher_update turned every error into "up to date" (offline, rate limit, non-semver tag). It now returns a Result: the failure is shown (an error toast on a manual click, the status line for an automatic check) and never claims Colony is current. The "up to date" toast only appears for a manual check, and boot no longer runs the check twice.
4. auto_check_updates is ON by default. A store that never checks leaves its badges invisible forever. The check is batched and costs zero API calls for apps pinned to a fixed tag.
5. The scan cache is finally read. It was written on every scan and never read back, so with the startup scan off every boot showed "0 apps". Boot now restores the cache when the scan is skipped.

Testing

cargo build and clippy --all-targets clean, 88 tests pass, boot smoke test OK (session restored, 3 repositories detected, no double launcher check).