Add GPU support

[royshil]

Using the GPU support from whisper.cpp will be crucial for real-time audio stream processing
It's already supported upstream for CUDA and CoreML: https://github.com/ggerganov/whisper.cpp
OpenVINO is also planned