fix(dashboard): drop the stale expert welcome while the next one loads

[JasonBuildAI]

Summary

Switching experts in Chat left the previous expert's welcome message and quick-prompt cards on screen — still clickable — until the new expert's /api/agents/{agent_id}/chat/welcome response arrived. Switching the UI language had the same effect.

useExpertChatWelcome() cancelled late responses, but it never cleared the state it had already loaded when a new request started, so Chat/index.tsx kept passing the stale values into WelcomeScreen for the newly selected expert.

The fix clears fetchedCards / welcomeSuffix immediately before starting the request, so the pending state is empty instead of the previous expert's content. The existing cancellation guard is unchanged: a late response from the expert that was switched away from is still ignored.

Fixes #1480

Target branch

• Base is develop (feature / fix — default)
• Base is main (release/* or hotfix/* only)

Type of change

• Bug fix
• New feature
• Breaking change
• Documentation
• Refactor / chore
• Release / hotfix

Test plan

Added dashboard/src/pages/Chat/hooks/useExpertQuickCards.test.ts — 4 cases: expert switch, language switch, late response ignored, no expert selected.

Before the fix — 2 failed, 2 passed:

$ cd dashboard && npx vitest run src/pages/Chat/hooks/useExpertQuickCards.test.ts
 FAIL  src/pages/Chat/hooks/useExpertQuickCards.test.ts > useExpertChatWelcome > drops the previous expert's welcome while the next request is pending
AssertionError: expected [ { title: 'Alpha任务', …(5) } ] to deeply equal []
 FAIL  src/pages/Chat/hooks/useExpertQuickCards.test.ts > useExpertChatWelcome > drops the previous welcome while the request for a new language is pending
AssertionError: expected [ { title: 'Alpha任务', …(5) } ] to deeply equal []
 Test Files  1 failed (1)
      Tests  2 failed | 2 passed (4)

After the fix — 4 passed:

$ cd dashboard && npx vitest run src/pages/Chat/hooks/useExpertQuickCards.test.ts
 ✓ src/pages/Chat/hooks/useExpertQuickCards.test.ts (4 tests) 109ms
 Test Files  1 passed (1)
      Tests  4 passed (4)

Other checks (Windows 11, Python 3.12, Node 22):

$ uv run ruff check src tests           -> All checks passed!
$ uv run ruff format --check src tests  -> 1168 files already formatted
$ uv run mypy src/octop                 -> Success: no issues found in 565 source files
$ cd dashboard && npx tsc -b            -> exit 0
$ cd dashboard && npm run build         -> ✓ built in 2m 43s

• make is not installed in this shell, so the four gate commands were run individually instead of make all.

• Full dashboard suite (npx vitest run): 10 failed | 1206 passed. All timeouts in unrelated files. Re-running those test files on a pristine develop checkout gives the identical 7 files / 4 tests failing (and the remaining 4 passing), so this change adds no new failures.

• Backend suite (uv run pytest -n auto -m "not live"): 53 failed | 4322 passed | 158 skipped. The same 53 fail on a pristine develop checkout (verified with a serial --lf re-run on both trees), so they are pre-existing on this Windows machine.

• npm run format:check flags 1105 files on this checkout because core.autocrlf=true checks everything out as CRLF; Prettier passes on the two files touched here.

• npm run lint reports 3 pre-existing errors in other files; none of them in the files touched here.

• make all passes locally (individual ruff / mypy / pytest commands; see above)

• Added/updated tests

Checklist

• Updated CHANGELOG.md (if user-facing)
• README / docs updated (if needed)

---

AI disclosure: this change was drafted with AI assistance at my request, under my direction and review.