When the PostgreSQL connection URL changes (e.g. pointing to a different database), local push watermarks become stale and the incremental push skips most sessions.
The fix would be to store a hash or identifier of the target PG database alongside the watermark, detect when it changes, and automatically clear watermarks to trigger a full re-push.
Current workaround: run `pg push --full` after re-pointing.
Tracked from the pg-sync refactor spec (out of scope for initial implementation).