/** * extractVideosStage — pre-extract source-video JPEG sequences, plus the * HDR color-space pre-detection that runs against the originals. * * The stage runs the existing video-frame extraction pipeline * (`extractAllVideoFrames`) but also probes BOTH videos and images for * native HDR color spaces before extraction (since extraction may convert * SDR → HDR). The HDR maps are returned so the downstream HDR auto-detect * block and the HDR composite path can identify which sources are natively * HDR vs. converted-SDR. * * Hard constraints preserved verbatim from the in-process renderer: * - `composition.audios` is mutated in place to add audio entries * auto-discovered from video files via ffprobe (preserves the * "video had audio, no explicit