--diarize flag no longer works with stereo input in latest release

[varlog11]

Environment:
whisper.cpp version: latest (1.7.5)
Platform: Debian 12, x86_64
Build flags: -D WHISPER_FFMPEG=yes -D GGML_CUDA=no
Test file: 2-channel stereo WAV with two clean channels

In version 1.7.4, the --diarize option correctly performed voice diarization on a stereo input file (splitting tracks by speaker/channel). After upgrading to the latest release, the same command no longer produces separate speaker outputs.

[danbev]

Could you provide a sample audio file that I can use to reproduce this, and also what model are you using?