fix(desktop): map branch targets through folded rows

[normaltank]

What does this PR do?

Fixes branch-from-message truncation when the clicked renderer bubble corresponds to a later source row inside a folded authoritative assistant message.

The hydrated transcript can fold several persisted assistant/tool rows into one bubble. The bubble keeps the first row as its top-level rowId, while each text part retains its own sourceRowId. selectBranchMessages() previously checked only the top-level ID, so a click on a later folded row could fail both mapping strategies and silently seed the child from the shorter renderer-local prefix.

This change maps the clicked bubble against every durable row represented by both bubbles (top-level rowId plus part sourceRowId). The existing same-role/text ordinal and local-prefix fallbacks remain unchanged.

Related Issue

Fixes #136193

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature (non-breaking change that adds functionality)
• 🔒 Security fix
• 📝 Documentation update
• ✅ Tests (adding or improving test coverage)
• ♻️ Refactor (no behavior change)
• 🎯 New skill (bundled or hub)

Changes Made

• Update apps/desktop/src/app/session/hooks/use-session-actions/utils.ts to match branch targets through folded part source rows.
• Add utils.branch.test.ts with a focused regression where the local clicked row is represented only by an authoritative part sourceRowId.
• Keep the oversized production file at its existing 2,609-line count.

How to Test

1. Before the implementation, run the new regression test: it fails by returning only the two-message local prefix instead of the four-message authoritative prefix.
2. Apply the implementation and run npx vitest run --project ui src/app/session/hooks/use-session-actions from apps/desktop.
3. Confirm all 15 files / 357 tests pass, including the folded-row regression.

Additional validation:

• npx tsc -p apps/desktop/tsconfig.json --noEmit
• ESLint on both changed files
• Prettier check on both changed files
• git diff --check
• python scripts/check --only footguns,shebangs,no-tmp,config-writers,os-marker-fakes,case-collisions,profile-archives,ruff --base origin/main (8 checks passed)

The full scripts/check health measurement was attempted, but its unchanged-base git grep scan timed out after 120 seconds in this sparse CI workspace; it did not report a code-health regression. The full Python suite was not run because this is a Desktop TypeScript-only change.

Checklist

Code

• I've read the Contributing Guide
• My commit message follows Conventional Commits
• I searched existing open and closed PRs, including fix(desktop): branch snapshots keep pre-compaction history, isolated from parent (salvage #76286) #86775, for duplicates
• This PR contains only changes related to this fix
• I've run pytest tests/ -q and all tests pass (not run; Desktop TypeScript-only change)
• Full python scripts/check passes (8 applicable checks pass; health measurement timed out as documented above)
• I've added a regression test
• Tested in the Linux CI workspace with Node 22

Documentation & Housekeeping

• Documentation update: N/A
• cli-config.yaml.example: N/A
• CONTRIBUTING.md / AGENTS.md: N/A
• Cross-platform impact considered: pure transcript-selection logic with no platform-specific behavior
• Tool descriptions/schemas: N/A

Screenshots / Logs

Focused behavior evidence:

• Before fix: 1 failed, 141 passed; received the two-message local prefix.
• After fix: 15 test files passed, 357 tests passed.

AI assistance disclosure: Codex investigated the issue, implemented the patch, added the regression test, and ran the validation listed above. No human code review is claimed.