Right now if we have an applyOps command with an 'alwaysUpsert: false' field, it will be applied with 'alwaysUpsert: false' both during steady state replication and in initial sync. This has the same idempotency problem as applying ordinary updates during initial sync, which is that the updated document may not be there because it might have been deleted on the sync source after the oplog entry. So we should ignore the applyOps alwaysUpsert field during initial sync.