We run snapshot reads on a restored node in magic restore tests twice – once with a timestamp prior to restore (it should fail) and then one at the stable timestamp set at the end of restore (it should succeed). In slow variants, it's possible restore completed significantly earlier than when snapshot read that should succeed is run. In that case, the snapshot will fail. We should extend the minSnapshotHistoryWindowInSeconds parameter on all restore tests to 1 hour to account for slow machines. Most restore tests should complete in under 5 minutes, so we shouldn't be taking up too much disk space.