You are correct -- native mode means that all of the control and response is done by the devices based on device links.
There are pros and cons of the "scene" vs. "program" question. As Michael points out, the scene offers the advantage of direct (native) action between the controller (the Remotelinc2) and the light. This eliminates any time delay and dependence on the ISY as the controller. The down side of the scene is that it offers no other means of adding intelligence to the activity. Using a program provides flexibility and allows more control of the results.
The original post indicated a desire to use non-toggle buttons to allow one button (A) to turn the light on and another button ( to turn the light off. Since the non-toggle mode for the Remotelinc2 appears to always provide an "ON" action, the easiest way to implement the second part of the desired control would to use a program looking for the "button B ON" to initiate the OFF command to the scene via the ISY. The button A control action (turn the light on) could either be performed by a similar program looking for the "button A ON" to initiate the ON command to the scene via the ISY or alternatively the ON could be performed by a scene linking the light with button A as the controller. I think that most people use the hybrid approach (scene + program) for this type of control, using the direct link (native mode) for ON (similar to allowing a motion detector to turn a light scene ON) and a program to manage the OFF control (for motion detectors, typically after some delay or in this case based on another input from the Remotelinc2).