Root Cause

The original implementation fetched poster artwork for the newly-selected item synchronously on the main render thread by shelling out to curl:

load_poster_art → resolve_art_path(file, true) → cached_remote_art_path(url, true) → blocking curl with a 10-second ceiling

Love2D is single-threaded. The --max-time 10 ceiling means the entire application freezes for up to 10 seconds per uncached poster while curl downloads the image. During this time:

• Render loop stalls (no frames drawn)
• Input events queue but are not processed
• User experience: pressing Down freezes the screen, then it jumps

This happens for every poster not already in data/image-cache/. The cache mitigates repeat visits to the same library position, but cold scrolls through a large library (Jellyfin Movies) trigger the fetch for every new selection.

The follow-up RG smoke test found two remaining blockers after fetches moved to a worker:

• The Jellyfin catalog was still fetched synchronously from love.load, so the app could show only local files after an empty response, then block for several seconds on a later startup while VOD titles merged.
• Cached Jellyfin posters were still full-size primary JPEGs and were decoded synchronously on the Love2D main thread when selection changed or grid cells rendered.

The apparent black-screen incident was also traced to R2 lock mode calling batocera-brightness dispoff; it was a visible-state problem rather than a Love2D crash.