Add DTW token timestamps

[obvirm]

jfk_final.mp4

& '.\whisper-cpp-dtw - Copy\build\bin\Release\whisper-cli.exe' -m .\whisper-cpp-dtw\models\ggml-base.en.bin -f sample_audio.wav --dtw base.en --no-flash-attn --max-len 1 --output-srt -of test_copy

• Add phoneme-based onset shift (vowels/plosives 150ms, consonants 80ms)
• Enforce adaptive minimum token duration based on text length
• Cap maximum duration to prevent stretching over silences
• Sync segment boundaries with corrected token timestamps

Improves DTW timestamp accuracy for word-level subtitle generation.