fix(web): adopt six upstream chat fixes for the composer bar, question rows and file links

[rynfar]

Problem

Six small web chat defects that upstream T3 Code fixed and Pylon still had (catch-up cycle for #865):

• The collapsed composer context bar flipped between labels and icons on every measure, because a middle-truncated branch label was measured by one element's scrollWidth and lost its tail.
• On desktop, the empty workspace and the "Couldn't start a new thread" screen had no drag region, so the window could not be moved.
• Reconnecting to a server whose version differs was presented as "Finishing an update" even when the server was simply offline.
• Question rows in the timeline led with a generic "User input submitted/requested" label instead of what the agent asked.
• Inline code such as z-ai/glm-5.3 or Qwen/Qwen2.5-Coder became a file chip.
• A link to the workspace root resolved to no relative path, went through basename lookup, opened a file surface and attempted a file read.

Fix

Source (upstream PR)	Disposition
29abbf9b424b122c52bb20dccffac68bee367c09 (#13555)	Adapted. Labels are measured by summing their text runs, capped at the motion span's max width.
eeea71a88108c375c04853f58c298a518a54acaf (#13713)	Adapted. Electron gets a titlebar drag strip on the no-projects hero and the draft-start error.
dd582dee36976ccfd2a896cf50c3ee957ca7b9be (#13083)	Adapted. The version-skew reconnect fold is removed; a reconnecting or offline server always shows the offline notice after the grace period.
916ec94f93da686303f878d506a760efd13506c1 (#12900)	Adopted (cherry-pick, conflicts resolved). Question rows lead with the question text; the answer stays as the trailing preview.
88fbc2cac77ff2bf0d406c44754ecd8014976f5f (#13909)	Adopted (clean cherry-pick). Version-suffix guard plus nine regression cases.
55ec55b1fb2e65f2736ff8420ae0373210c1911e (#12449)	Adopted (clean cherry-pick). Workspace root normalizes to ., which opens the singleton files explorer and suppresses the file read.

Pylon adaptations

• Composer bar: only the label text measurement changed. Pylon's hosted resting-composer controls measurement and has-content tracking in useLabelsOverflow are untouched.
• Offline notice: Pylon's description line, disabled "Reconnecting..." button and Connections action are kept; only the "Finishing an update" branch and its suppression of the update banner are removed. Update banners and dialogs keep their Pylon copy.
• Question rows: Pylon's expanded "Command" heading (displayText) and user-input.answer-submitted folding are kept. Upstream's AgentSpawnRow context in the conflict is not part of this source and was not brought in. The new getQuestionTextPreview tests are appended to Pylon's existing userInput.test.ts.
• Empty workspace: Pylon's hero and error classes are kept; only the drag strip is added.

Exclusions

• Upstream's electron={isElectron} on the hosted-static onboarding header belongs to a different upstream commit and is not included.
• Mobile is not changed. It consumes the shared inline-code heuristic (apps/mobile/modules/t3-markdown-text/src/markdownLinks.ts) and so gets the model-id guard; it does not call workspaceRelativeFilePath.

Verification

• vp test run on packages/client-runtime/src/markdownLinks.test.ts, packages/client-runtime/src/work-log/userInput.test.ts, apps/web/src/components/chat/MessagesTimeline.test.tsx, apps/web/src/components/files/FilePreviewPanel.test.ts, apps/web/src/markdown-links.test.ts, apps/web/src/rightPanelStore.test.ts, apps/web/src/workspaceBasenameLookup.test.ts, apps/mobile/src/lib/markdownLinks.test.ts: 8 files, 329 tests passed (rerun after rebase).
• vp run -F @t3tools/web typecheck and vp run -F @t3tools/client-runtime typecheck: both ran and exited 0 with no errors (run before the final rebase, which only brought in server commits).
• vp lint on the 18 changed files: 0 errors; the warnings reported are on lines this PR does not touch.
• vp fmt --check on the 18 changed files: clean.
• The composer measurement, drag strip and offline notice changes have no unit tests upstream or here.

Limits

No rendered-client pass was done: no browser, desktop or simulator run. The composer measurement, drag region and offline notice are verified by typecheck and review only.

Refs #865

Model: Claude Opus 5.5. Harness: Claude Code in Pylon. Workflow: review-t3-upstream skill.

---

^{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	14.0 KiB	14.0 KiB	−4 B (−0.0%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.3 KiB	7.3 KiB	−4 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.7 KiB	6.7 KiB	0 B (0.0%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	58.0 KiB	58.0 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	9	9	0 (0.0%)	21	✅
Claude	Total thread wire	14.0 KiB	14.0 KiB	−24 B (−0.2%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.3 KiB	7.3 KiB	+2 B (+0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.8 KiB	6.7 KiB	−26 B (−0.4%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	58.9 KiB	58.9 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	9	9	0 (0.0%)	21	✅

Baseline: da77592 · PR result: 20662ac · 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: 115.7 KiB
• Claude decoded thread snapshot: 116.4 KiB

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