Fix namespace in examples

sycl/doc/SYCLGraphUsageGuide.md

@@ -50,6 +50,8 @@ placed into a `host-task` node which the kernel (or other operation) depends on,
 like so:
 
 ```c++
+Graph.begin_recording(Queue);
+
 sycl::event HostWorkEvent = Queue.submit([&](sycl::handler& CGH){
     CGH.host_task([=](){
         do_some_host_work();
@@ -60,6 +62,8 @@ Queue.submit([&](sycl::handler& CGH){
     CGH.depends_on(HostWorkEvent);
     CGH.parallel_for(...);
 });
+
+Graph.end_recording(Queue);
 ```
 
 ### Graph Execution Concurrency
@@ -73,7 +77,9 @@ synchronization. Consider this example (it uses whole graph update for
 simplicity):
 
 ```c++
-sycl::command_graph<graph_state::executable> ExecutableGraph
+namespace sycl_ext = sycl::ext::oneapi::experimental;
+
+sycl_ext::command_graph<graph_state::executable> ExecutableGraph
    = ModifiableGraph.finalize();
 
 Queue.ext_oneapi_graph(ExecutableGraph);
@@ -92,9 +98,11 @@ accessing the same `sycl::buffer`) for updating, and potentially increasing
 device occupancy (if the device supports it). Modifying the example from above:
 
 ```c++
-sycl::command_graph<graph_state::executable> ExecutableGraph
+namespace sycl_ext = sycl::ext::oneapi::experimental;
+
+sycl_ext::command_graph<graph_state::executable> ExecutableGraph
    = ModifiableGraph.finalize();
-sycl::command_graph<graph_state::executable> ExecutableGraphOtherParams
+sycl_ext::command_graph<graph_state::executable> ExecutableGraphOtherParams
    = ModifiableGraphWithNewParams.finalize();
 
 Queue.ext_oneapi_graph(ExecutableGraph);