af0796c835
Two new floating browsers, both fed only by static data — the disc for screens, a savedata file for saves — following the existing Game Data / Ships pattern (request event → worker thread → IsoLoaderMsg → draw system). UI Screens Pick a screen pak and one of its builds (a build = one context × language variant; GP_PAUSE_MENU holds six) and see the screen reassembled at 1280×720 beside the element list it was derived from, so a wrong placement shows up as a wrong picture and the row next to it says why. Per-element visibility toggles, focused-state and loop-animation toggles, and an inspector showing each element's sprite, parent link, kind flags, pivot and keyframe count. Compositing decodes ~30 T8aD sprites, which is far too slow for the UI thread, so it runs on a worker under the same generation gate the XPR path uses — a toggle mid-decode discards the stale result rather than racing it. The panel deliberately surfaces two things it would be easy to hide: sprites that named a child the bundle does not hold, and builds recovered through the .rat fallback (where elements without a record are simply absent). Save File Opens a savedata file (it lives in the emulator's content tree, not on the disc, so this is a file dialog rather than a disc path). Shows the container, the GHAD block with every field coloured by its actual confidence, the Arsenal develop blob, the per-stage SHAB records, and the header summary. Three things the panel states rather than glosses: the byte-identical round-trip, which is the check the whole layout rests on; that SHAB records are per-stage results and NOT the UI's save slots; and that the header summary is what the Details panel reads, so a payload edit that leaves it stale proves nothing about the field that was edited. Fields tested and refuted (+36, +56) are shown as refuted rather than dropped. Also gates draw_game_data_ui / draw_ships_ui at their registration site. Both are #[cfg(not(wasm32))] but were registered unconditionally, so the wasm leg of `just ci` failed on an undefined name. NOTE: that leg still does not build, for a pre-existing and unrelated reason — the workspace pins tokio with features = ["full"], which pulls mio, which refuses to compile for wasm32. Fixing that means restructuring tokio's features per target and is left alone here. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>