1.2 GiB of MP3 voice data won't fit in the per-asset Gitea ceiling and
would make the pack 25x larger. The small AI_VoiceOver framework on its
own is useless without the data, so excluding the whole repo is the
right call. Track distribution of the voice data separately.