Avoid retrieving context data for every single node which can be
expensive when there are thousands of nodes. In the "Mouse House"
test file I observed a 13% improvement in drawing timings.

@@ -344,3 +354,2 @@
     const bNode &node = *nodes[i];
-    std::string block_name = "node_" + std::string(node.name);
-    uiBlock *block = UI_block_begin(&C, CTX_wm_region(&C), std::move(block_name), UI_EMBOSS);
+    uiBlock *block = UI_block_begin(&C, scene, window, region, std::string(node.name), UI_EMBOSS);

@@ -4799,6 +4850,7 @@ static void node_draw_nodetree(const bContext &C,
                                Span<uiBlock *> blocks,
                                bNodeInstanceKey parent_key)
 {
+  SCOPED_TIMER_AVERAGED(__func__);