transcribe.cpp enables ggml speech-to-text for 16 model families

This C++ implementation provides speech-to-text inference using the ggml framework. It supports over 16 different model families for high-performance local audio transcription.