Right now we require separate files for each IDS_* message. msw suggested that we can just embed them as another tag inside grd/grdp files.

The translation pipeline could then extract hashes from these files and download images when neeeded.