Sanitized whisper.cpp parallel-timestamp evidence

Environment:
  Windows 11 x64
  Lemonade SDK whisper.cpp-rocm v1.8.4 whisper-cli
  AMD Radeon RX 9070 XT, Vulkan backend
  ggml-base.bin
  Input duration: 3630.8 seconds, decoded mono 16 kHz

P4 output:
  whisper_full_parallel: the audio has been split into 4 chunks at the following times:
  whisper_full_parallel: split 1 - 00:15:07.690
  whisper_full_parallel: split 2 - 00:-14:-28.-960
  whisper_full_parallel: split 3 - 00:00:38.730

P4 example segment offsets:
  from: 1815370 ms
  to:   -859960 ms

P2 output:
  whisper_full_parallel: the audio has been split into 2 chunks at the following times:
  whisper_full_parallel: split 1 - 00:-14:-28.-960

Attempt results on the same input before downstream timestamp repair:
  P4: inference completed in 16.70 seconds; 67 invalid segment records
  P2: inference completed in 23.73 seconds; 65 invalid segment records
  P1: inference completed in 40.20 seconds; valid timestamps

The transcript text and source identity have been omitted.