Add save health, durable retries, and a clearer terminal dashboard

[Vasanthdev2004]

Summary

This turns save health into a visible, actionable part of the existing CLI/dashboard and makes offline recovery restart-safe.

• Add shared protection states, last successful upload timestamps, actionable failures, and aggregate tray health.
• Persist upload intent atomically with snapshots; retry with backoff in both watchers, or manually via gsg retry / dashboard U.
• Retain queued snapshots through local retention. Block damaged snapshots without deleting their files/history, and allow explicit manual revalidation after repair.
• Serialize backup, snapshot registration, upload, restore, and purge with a separate reentrant writer lock. A missed automatic close backup is persisted and retried before uploads/cloud restores.
• Immediately protect games discovered during rescans and honor current pause/remove/destination settings before retries.
• Polish gsg status for 80 columns and the Textual dashboard: search, compact/wide layouts, readable timestamps and labels, explicit Safety copies, scrollable health, collapsible activity, and clearer keyboard affordances.

Dashboard

The screenshot uses a synthetic example library; no personal saves or cloud credentials are included.

Verification

• Full suite: 394 passed on Windows / Python 3.11, including actual rclone CAS round trips.
• Ruff and strict mypy: clean.
• Runtime dependency declaration check: clean; no new dependency.
• Textual keyboard/search/restore interactions, stale cloud results, empty/error states, and 80×24 / 140×40 layouts exercised.
• All 27 dashboard tests also passed in an isolated environment with Textual 0.80.0, the minimum supported version.
• Independent safety review found and fixed deferred-close backup loss, stale retry settings, and endless damaged-snapshot retries.

Operational notes

• Restart older watcher processes after upgrading so they use the new writer lock.
• Automatic retries require gsg auto or gsg watch to be running.
• Queued/blocked snapshots can temporarily exceed max_versions; preserving pending saves takes priority over disk retention.
• Blocked historical snapshots need their original file recovered and an explicit manual retry; new healthy backups can still upload.
• Existing records migrate without inventing upload intent or historical upload timestamps. Protected means a completed local-recorded upload, not a fresh remote availability check.
• No version bump, release, or automatic merge.