Neural Population Learning beyond Symmetric Zero-sum Games