Voxa.Audio.Diarization

Speaker diarization for the Voxa pipeline (VLS-005) — "who spoke when", filling TranscriptionFrame.SpeakerId. Defines the ISpeakerSegmentation / ISpeakerEmbedding / IDiarizer seams and a pure-C# DiarizationPipeline (constrained agglomerative clustering by cosine distance — no ML runtime in the orchestration). Ships the seams + clustering; the reference ONNX impls (Pyannote segmentation + WeSpeaker embedding) are a separate opt-in package.