PT-4179: fix Storybook websocket-timeout crash for renderer stories

[katherinejensen00]

Summary

Renderer app components (the startup-wizard shell/steps, plus dialogs and overlays) import useLocalizedStrings from @renderer/hooks/papi-hooks and otherwise reach the network service. Storybook has no PAPI backend, so the renderer RpcClient's connection attempt to ws://localhost:<port> never settles and its AsyncVariable('websocket connected') rejects unhandled after ~10s — Timeout reached when waiting for websocket connected to settle — which the dev overlay surfaces as a crash on every startup-wizard story. Storybook previously stubbed only @papi/* imports, never the @renderer/@shared paths these renderer components actually use.

This is split out of the PT-4179 sync-progress work as a standalone fix because it is shared Storybook infra, not step-specific: it also unblocks the sibling wizard PRs (PT-4176 language, PT-4177 registration, PT-4178 sync-consent), which all render FirstRunShell and hit the same crash. Landing it on main lets those PRs inherit it on rebase.

Why review this

Supports the startup-wizard epic: the wizard step stories currently crash in Storybook, blocking visual review of PT-4176/4177/4178/4179. This restores them (and other renderer-component stories) to a working, backend-free state.

Changes

• .storybook/papi-stubs/renderer-papi-hooks.ts (new) — re-exports the real renderer hooks but overrides only useLocalizedStrings to resolve real English strings synchronously (no network).
• .storybook/papi-stubs/rpc-handler.factory.ts (new) — inert RPC handler so networkService.initialize() succeeds with no socket and no timer; no connection is ever attempted, so nothing rejects.
• .storybook/main.ts — wires both via NormalModuleReplacementPlugin.
• src/renderer/components/notification-display.test.tsx — add afterEach(() => { act(() => { toast.dismiss(); }); }) to cancel Sonner's pending auto-dismiss timers before jsdom tears down. Without this, a 4 s Sonner timer fires after the test environment is torn down, causing window is not defined to surface as an unhandled Vitest error that exits the whole test run with code 1 (all 1191 tests pass, but the process fails). This is a pre-existing race condition introduced in the PR that added Sonner (PT-4193/PT-4193: Add secondary notification action, position, and dismissible #2561); fixing it here unblocks CI on this PR.

Both Storybook stub files use NormalModuleReplacementPlugin, not resolve.alias: the base renderer webpack config resolves @renderer/@shared via TsconfigPathsPlugin, which wins over resolve.alias (that is why the existing @papi/* aliases work, but a @renderer/... alias would be silently ignored). The plugin rewrites the request in beforeResolve, before TsconfigPaths runs.

AI Involvement

AI-assisted. Claude diagnosed the root cause (traced the unhandled rejection from useLocalizedStrings → dataProviderService.get → networkService.initialize → RpcClient), wrote the two stubs and the main.ts wiring, and verified the result. Claude also diagnosed the CI test failure (Sonner timer leak in notification-display.test.tsx) and wrote the afterEach fix. The author reviewed the diagnosis, the stub approach, and the diffs.

Testing

• npm run storybook:build passes (the @papi-free hard gate) with both stubs.
• Rendered SyncProgressStep (Default, All Done) and FirstRunShell (Language) against the production build: real English strings render and zero unhandled rejections (was: Timeout reached when waiting for websocket connected to settle).
• Author confirmed the crash is gone in local dev Storybook.
• notification-display.test.tsx locally: timer leak no longer surfaces as an unhandled error.

Risk Level

Low — Storybook config only (no renderer production code changed); NormalModuleReplacementPlugin rewrites only the two targeted module paths. The notification-display test fix is additive (adds afterEach cleanup only). If either stub is wrong, storybook:build fails cleanly rather than silently.

---

This change is 

[jolierabideau]

Looks good to me, other than the lint failures!

@jolierabideau reviewed 4 files and all commit messages, and made 2 comments.
Reviewable status: all files reviewed, 1 unresolved discussion (waiting on katherinejensen00).

---

.storybook/papi-stubs/renderer-papi-hooks.ts line 2 at r2 (raw file):

/**
 * Storybook stub for the renderer PAPI-hooks barrel (`@renderer/hooks/papi-hooks`).

NIT From Claude

renderer-papi-hooks.ts re-exports the 11 non-localization hooks from src/renderer/hooks/papi-hooks/index.ts by hand. I verified they match exactly today. But since the $-anchored replacement redirects the whole barrel to this stub, a hook added to the real barrel later would be missing here, and any story importing it would fail to resolve. It'd surface as a Storybook build error (not a runtime bug), so it's low-severity — but a one-line comment on the real barrel pointing at this stub as the sync target (or a type-level assertion that the stub covers the barrel) would make the coupling discoverable.

[jolierabideau]

@jolierabideau reviewed 2 files and all commit messages, made 1 comment, and resolved 1 discussion.
Reviewable status: complete! all files reviewed, all discussions resolved.