Mesa TensorFlow Lite external delegate

 TensorFlow Lite delegate which can make use of NPUs to accelerate ML
 inference. It is implemented in the form of a external delegate, a shared
 library which the TensorFlow Lite runtime can load at startup.