fix(web): fold interim turn responses

[rynfar]

A settled turn kept both its first and its terminal assistant message visible, so a
turn that opened with a long interim response left that response permanently
expanded above the "Worked for ..." row. The duration row then sat between two
messages instead of leading into the answer.

Settled turns now keep only the terminal assistant message. Everything before it
folds behind the duration row, so the fold reads as "worked for N, here is the
result". Agent-spawn CTA rows still never fold.

Adopted from upstream pingdotgg/t3code#8828 (17c48f7fc), cherry-picked clean.

This is a deliberate behaviour change, not a bug fix — it changes what is
visible on every settled turn. Called out because it is worth a look in a real
client before it lands, not just a green test run.

Verified: vp test run --dir apps/web apps/web/src/components/chat/MessagesTimeline.logic.test.ts apps/web/src/components/chat/MessagesTimeline.test.tsx
(90 passed), vp run -F @t3tools/web typecheck (exit 0, no errors).

Reviewed and integrated with 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.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.6 KiB	13.5 KiB	−32 B (−0.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	0 B (0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.6 KiB	6.6 KiB	−32 B (−0.5%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	57.2 KiB	57.1 KiB	−44 B (−0.1%)	66.4 KiB	✅
Codex	Live turn messages	10	9	−1 (−10.0%)	21	✅
Claude	Total thread wire	13.6 KiB	13.5 KiB	−63 B (−0.5%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	−12 B (−0.2%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.6 KiB	6.6 KiB	−51 B (−0.8%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	58.1 KiB	58.0 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	10	9	−1 (−10.0%)	21	✅

Baseline: 9ade80d · PR result: 97aaed8 · 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: 109.5 KiB
• Claude decoded thread snapshot: 110.2 KiB

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

[rynfar]

Adversarial review confirmed the mechanics are clean — firstAssistantEntry had no remaining references, group.terminalEntry is null-safe, the zero-hidden short circuit still holds, and the turn-diff "changed files" chip stays on the terminal message rather than folding away. Two findings.

Fixed: a turn that ends with no text folded away its own answer. With the first-message exemption gone, a settled turn whose terminal assistant message is empty renders as Worked for 22s above (empty response), with the substantive response hidden and no item count on the fold row. The comment upstream deleted named exactly this case and no substitute guard replaced it. The first assistant message now stays visible when the terminal message carries no text.

I deliberately kept this narrow. My first attempt also exempted short terminal messages (under 80 chars), which broke upstream's own fixture where the turn ends "Verification finished." — that is upstream's product call about terse closers, not a defect, so a terse-but-non-empty ending still folds. Only the empty case is guarded. Regression test added.

Not fixed — belongs with the mobile sequence. apps/mobile/src/lib/threadActivity.ts:1261 still exempts the first assistant message from folding, so web and mobile now render the same settled turn differently. The two fold implementations are duplicated and drifting. I'm holding this for the mobile batch, where #8793 already moves work-log presentation into client-runtime — that is the right moment to converge the two rather than patch mobile in a web PR.

Re-verified: MessagesTimeline.logic.test.ts 47 passed, MessagesTimeline.test.tsx 42 passed, @t3tools/web typecheck clean.