refactor: optimize CallbackExecutor for better performance

- Execute plugin_callback directly instead of wrapping in CallbackPipeline
- Makes plugin callback priority more explicit
- Fixes incorrect lambda in docstring example
- Reduces unnecessary overhead for single callback execution

Addresses feedback from gemini-code-assist bot review (round 2)

Author: jaywang172

src/google/adk/agents/callback_pipeline.py

@@ -228,18 +228,20 @@ async def execute_with_plugins(
       First non-None result from plugin or agent callbacks, or None.
 
     Example:
+      >>> # Assuming `plugin_manager` is an instance available on the
+      >>> # context `ctx`
       >>> result = await CallbackExecutor.execute_with_plugins(
-      ...     plugin_callback=lambda: plugin_manager.run_before_model_callback(
-      ...         callback_context=ctx,
-      ...         llm_request=request,
-      ...     ),
+      ...     plugin_callback=ctx.plugin_manager.run_before_model_callback,
       ...     agent_callbacks=normalize_callbacks(agent.before_model_callback),
       ...     callback_context=ctx,
       ...     llm_request=request,
       ... )
     """
     # Step 1: Execute plugin callback (priority)
-    result = await CallbackPipeline([plugin_callback]).execute(*args, **kwargs)
+    result = plugin_callback(*args, **kwargs)
+    if inspect.isawaitable(result):
+      result = await result
+
     if result is not None:
       return result