fix(web): stop fork markers rendering as UI text, and remember the right panel's shape

[tusharbhardwaj-bk]

Problem

Two things, both reported from expbkt3 after the upstream merge.

1. Fork marker comments were being rendered into the UI. Four // T3-CUSTOM(expbkt3): ... comments landed in JSX children position inside the traits chip, so React rendered them as literal text next to the model picker — visible in the composer, in Settings → General (where the stray text also wrecked the "Default agent model" row layout), and in every project's new-thread defaults. My mechanical marker pass during the merge put them there; markers in attribute position and in plain TypeScript are fine, children position is not.

2. The right panel forgot how you like to work. Upstream keeps "is the right panel maximized" as a per-thread key in component state, so opening a plan full-screen in one thread and then opening a plan in another gave you side-by-side again — and a reload always reset it.

How

Removes the four comments, and adds apps/web/src/forkMarkerPlacement.test.ts: it walks every .tsx under apps/web/src and fails on any T3-CUSTOM marker sitting in JSX children, while allowing the legitimate attribute-position ones. Verified it catches the regression by re-introducing one deliberately — it failed and named the exact line.

For the panel, the maximized choice moves into local storage behind a fork-owned hook (rightPanelLayoutPreference.ts), so every surface — plan, browser, files — opens in the shape you last chose, and closing a panel no longer discards it. Width needed no work: PreviewPanelShell already persists it globally under t3code:preview-panel-width.

Kept deliberately cheap to re-merge: one new fork-owned module plus four marked lines in ChatView.tsx (the state declaration, the derivation, the toggle, and the close path), rather than logic threaded through the component.

Verification

apps/web typechecks clean; vp check clean on the touched files. New tests pass (4 preference + 1 placement guard), as do ChatView.logic, TraitsPicker, Sidebar.logic and session-logic (263 tests). Fork marker check passes.

Model: Claude Opus 5 in Claude Code.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

ℹ️ No successful main baseline artifact is available yet. This run establishes the initial measurement.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	—	13.6 KiB	—	15.1 KiB	✅
Codex	Thread snapshot wire	—	7.2 KiB	—	7.3 KiB	✅
Codex	Live turn WebSocket wire	—	6.4 KiB	—	7.8 KiB	✅
Codex	Live turn WebSocket decoded	—	55.7 KiB	—	66.4 KiB	✅
Codex	Live turn messages	—	11	—	21	✅
Claude	Total thread wire	—	13.6 KiB	—	15.1 KiB	✅
Claude	Thread snapshot wire	—	7.2 KiB	—	7.3 KiB	✅
Claude	Live turn WebSocket wire	—	6.4 KiB	—	7.8 KiB	✅
Claude	Live turn WebSocket decoded	—	56.5 KiB	—	66.4 KiB	✅
Claude	Live turn messages	—	11	—	21	✅

Baseline: unavailable · PR result: ec3bb06 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 110.4 KiB
• Claude decoded thread snapshot: 111.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.