Reachy Mini goes fully local

You can now run the entire conversational AI for the Reachy Mini robot completely on your local machine, ensuring privacy and eliminating reliance on cloud services. This is achieved through a cascaded pipeline that processes voice activity detection, speech-to-text, LLM inference, and text-to-speech generation. The setup involves running a local LLM server, such as llama.cpp with a Gemma model, and connecting it to the `speech-to-speech` engine. This local approach not only removes API costs but also grants you full control to customize the pipeline by swapping in different models for each stage.