feat: add transcript endpoint

[mudler]

Part of: #138

This is a quick wire to whisper.cpp following loosely the work already in https://github.com/go-skynet/whisper, wires https://github.com/ggerganov/whisper.cpp with go and uses ffmpeg for converting input audios to wav.

It adds transcription endpoint to LocalAI - however:

• It does not output the format indicated by the user (this is a first implementation round, lots to refine), always defaults to { "text" : " ... " }
• It shells out to ffmpeg to convert the audio input (it is not added to the image, separate images must be arranged for this)
• lots to improve