Snap video cuts to sentence boundaries with whisper word timestamps, not silencedetect

ffmpeg silencedetect at about -38dB with d=0.2 also flags breaths between words, so cut points placed in its gaps can split sentences. The user hears this as the voice cutting out. Get word timestamps with whisper.cpp (whisper-cli -m ggml-base.en.bin -ml 1 -sow), and make each cut remove or keep whole sentences. Trust silencedetect over whisper inside long silences, because whisper smears word times across them. Separately, Teams recordings contain mic dropouts: digital silence that shows up at silencedetect -80dB, often exactly 13.3s long and mid-sentence. That audio is unrecoverable, and the user may blame the edit for it. Detect these at -80dB and trim the dead air instead. In FilmCraft, ripple-trimming one track of an unlinked V/A pair under sync-lock leaves a gap on the other track. Fill it with a regular (non-ripple) trim, not closeGap, which refuses to break sync.