[dag] Fix the workflow doc code build (ray-project#37249)

Previously this code was relying on an implementation detail (DAG nodes not being serializable) to illustrate the point about static DAGs. Serializability was enabled in ray-project#37198.

I've updated the code sample to show that the DAG must be constructed statically and returning nodes from other nodes will be evaluated at runtime.

doc/source/workflows/basics.rst

@@ -308,30 +308,31 @@ Idempotent workflow:
 Dynamic workflows
 -----------------
 
-Workflow tasks can be dynamically created in the runtime. In theory, Ray DAG is
-static which means a DAG node can't be returned in a DAG node. For example, the
-following code is invalid:
+Ray DAGs are static -- returning a node from another node isn't a valid way to
+construct a graph. For example, the following code prints a DAG
+node, not the output of `bar`:
 
 .. testcode::
 
  @ray.remote
- def bar(): ...
+ def bar():
+ print("Hello from bar!")
 
  @ray.remote
  def foo():
- return bar.bind() # This is invalid since Ray DAG is static
+ # This is evaluated at runtime, not in DAG construction.
+ return bar.bind()
 
- try:
- ray.get(foo.bind().execute()) # This will error
- except ray.exceptions.RayTaskError:
- print("Ray DAG is static")
+ # Executing `foo` returns the `bar` DAG node, *not* its result.
+ print("Output of foo DAG:", type(ray.get(foo.bind().execute())))
 
 .. testoutput::
 
- Ray DAG is static
+ Output of foo DAG: <class 'ray.dag.function_node.FunctionNode'>
+
 
-Workflow introduces a utility function called ``workflow.continuation`` which
-makes Ray DAG node can return a DAG in the runtime:
+To enable dynamically executing DAG nodes at runtime, workflows introduces a utility
+function called ``workflow.continuation``:
 
 .. testcode::