Xiaomi Released MiMo-Audio, a 7B Speech Language Model Trained on 100M+ Hours with High-Fidelity Discrete Tokens