What should we add?

We need a qk_dag_topological_op_nodes function to enumerate all operation nodes (their indices) in the DAG.

The C API demo code imagines a function like this:

void qk_dag_topological_op_nodes(const QkDag *, uint32_t *order);

which expects the caller to allocate their own buffer for order, using an allocation of size qk_dag_num_instructions. Then, qk_dag_topological_op_nodes would just fill that buffer eagerly, rather than us supplying any sort of iterator.

If we'd rather provide an iterator, we should be consistent between this function and others that enumerate the graph.