Add support for Gemma 3N

Gemma 3N explicitly has better efficiency for on-device inference, making it perfect to be used in the context of executorch.

I can help if I can be pointed in the right direction to start adding support, and perhaps someone to speak to with familiarity on the project. @guangy10 I'm seeing how much work you put into this and executorch, would be happy to help lessen the load and contribute!