Improve checkpointing for background sync

[Okuro3499]

Problem

After a full sync completes, five large tables — ratings, courses_progress, submissions, login_activities, and team_activities — are synced via a background coroutine which processes each sequentially by paginated batches (with copyToRealmOrUpdate).

There are two key failure modes:

1. No checkpoint persistence

   • Pagination skip offset (tracking sync progress) is only held in memory.
   • If the app is killed, force-stopped, or OOMs mid-way, or if network is lost, the next sync starts from skip=0 for the affected table.
   • On large installs, this can require re-downloading tens of thousands of rows on every interruption.

2. No per-session progress on network loss

   • If network goes down during background sync, the current offset is discarded.
   • On restoration, the user must trigger a new full sync from scratch.
   • Worse, in some cases (e.g. HTTP 503 from the server), code paths delete any partial progress, so even transient errors erase checkpoints.