WavShape: Information-Theoretic Speech Representation Learning for Fair and Privacy-Aware Audio Processing