[tiensonqin]

Summary

Lui_app.model app now reads Signal.state_pending before the published model in both application drivers. Reentrant reducers can queue A → B → A before signal publication; imperative async context fences must see the accumulated revision rather than the previous published snapshot. The public signature and reactive batching remain unchanged; reading does not flush rendering or patches.

A retained regression covers plain, extension-enabled, reloadable and reloadable-extension constructors: two queued reductions are immediately readable without emitting patches, and explicit flush preserves the result. The migration guide distinguishes authoritative reads from borrowed published view signals.

Validation: dune build @all; 236 core tests; 10 schema/tooling tests; headless dune runtest with LD_LIBRARY_PATH=$HOME/.local/lib SDL_VIDEODRIVER=dummy. The default environment first hit an outdated SDL symbol and then unavailable OpenGL bindings; the headless run uses the existing tests' GL-unavailable path, not GL pixel validation. No GUI or application parity acceptance is claimed.

Link to Devin session: https://app.devin.ai/sessions/4a14d15bc6d44d18a9254cd088d5cc5e
Open in Devin Desktop: https://app.devin.ai/desktop/session/4a14d15bc6d44d18a9254cd088d5cc5e?variant=devin
Requested by: @tiensonqin

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring