Whisper Model moved to CoreML

CPU usage is quite high on current whisper model. It's functional for evaluation, but the battery usage isn't shipable. Moving the model to CoreML would help a ton, offloading work from CPU and improving efficiency. Accelerate framework is using GPU/ATX/SIMD but not ANE.

This project has some work, but doesn't appear complete: https://github.com/vade/OpenAI-Whisper-CoreML