Audio import collapses multitrack and multichannel sources to mono

[predict-woo]

Summary

Importing split-channel or multitrack audio discards its existing channel separation before diarization.

Steps to reproduce

1. Import a recording with microphone and system audio on separate tracks or channels.
2. Inspect the generated audio.webm or its channel layout.
3. Compare speaker attribution with the source separation.

Actual behavior

The generated recording is mono, with every source track/channel mixed into one waveform before diarization.

Expected behavior

Existing source separation should be preserved or used during speaker attribution instead of being unconditionally discarded.

Technical context

MediaFileDecoder passes every audio track to one AVAssetReaderAudioMixOutput and always requests AVNumberOfChannelsKey: 1. There is no branch based on source track or channel count.

Reported by Joey Virrueta.