[cycsmail]

Videos with GPS in their embedded XMP had no position (#1260). ffprobe only exposes that XMP with -export_xmp 1, and up to ffprobe 4.3 the flag aborts the probe for any container whose demuxer doesn't know it (4.4+ only warns). So loadVideoMetadata now runs a second probe with the flag when the first one reports the mov/mp4 demuxer, and passes the XMP through the existing mapGPS, same parsing as the photo branch and XMP sidecars (sidecars still win). That is one extra ffprobe run per mp4/mov at index time. Only GPS is taken from the embedded XMP on purpose: mapping everything would also change the creation dates of videos that carry Adobe XMP.

Added a test that appends an XMP uuid box with GPS to a copy of video.mp4 at runtime (no new binary fixture); it failed before the fix. The MetadataLoader, DiskManagerWorker and ffmpeg video worker specs pass with it (120 passing).

Two notes. It reads the standard exif:GPSLatitude/exif:GPSLongitude XMP tags (what exiftool writes); the XMP-crd namespace from the issue is non-standard and I had no sample for it. And the folder overview map is fed through the photosOnly pipe, so after a reindex the video shows its position in the lightbox info panel and mini map, but not yet on the overview map. Happy to add videos there in a follow-up if you want them.