fix(ui): preserve question queue order when upserting duplicate requests

[JDis03]

Summary

• Fixes [Bug]: questions do not open and garble leading to deadlock "Waiting for earlier responses" #448 - Questions deadlock with "Waiting for earlier responses" when a question arrives first as a global entry and later resolves to a tool part with a newer timestamp.

Root Cause

upsertQuestion replaced the entire queue entry on duplicate request id, losing the original enqueuedAt value. If question-1 arrived at t=1000 (global), then question-2 at t=1500, and question-1 resolved to a tool part at t=2000, the newer enqueuedAt (2000) would place question-1 behind question-2 in the sorted queue, breaking active interruption ordering.

Fix

• Add mergeQuestionEntry mirroring mergePermissionEntry behavior
• Preserve original enqueuedAt using Math.min(existing, new)
• Keep newer messageId/partId when resolving from global to tool part
• Clean up stale byMessage entries for the same request id
• Sort queue by enqueuedAt and recalculate active after each upsert

Validation

• npm run typecheck --workspace @codenomad/ui
• node --import tsx --test packages/ui/src/stores/message-v2/instance-store.test.ts
• Added 3 regression tests covering duplicate resolution and ordering

[github-actions]

PR builds are available as GitHub Actions artifacts:

https://github.com/NeuralNomadsAI/CodeNomad/actions/runs/26034514624

Artifacts expire in 7 days.
Artifacts:

• pr-479-66367454adf9ba9c7603ea45be110e1e0ed1fe92-tauri-macos
• pr-479-66367454adf9ba9c7603ea45be110e1e0ed1fe92-electron-macos
• pr-479-66367454adf9ba9c7603ea45be110e1e0ed1fe92-tauri-linux
• pr-479-66367454adf9ba9c7603ea45be110e1e0ed1fe92-tauri-windows
• pr-479-66367454adf9ba9c7603ea45be110e1e0ed1fe92-tauri-macos-arm64
• pr-479-66367454adf9ba9c7603ea45be110e1e0ed1fe92-electron-windows
• pr-479-66367454adf9ba9c7603ea45be110e1e0ed1fe92-electron-linux

[github-actions]

PR builds are available as GitHub Actions artifacts:

https://github.com/NeuralNomadsAI/CodeNomad/actions/runs/26035402900

Artifacts expire in 7 days.
Artifacts:

• pr-479-4dc05d3773ea6e82137836637d00b9b89859471e-tauri-macos
• pr-479-4dc05d3773ea6e82137836637d00b9b89859471e-tauri-linux
• pr-479-4dc05d3773ea6e82137836637d00b9b89859471e-tauri-windows
• pr-479-4dc05d3773ea6e82137836637d00b9b89859471e-electron-macos
• pr-479-4dc05d3773ea6e82137836637d00b9b89859471e-tauri-macos-arm64
• pr-479-4dc05d3773ea6e82137836637d00b9b89859471e-electron-windows
• pr-479-4dc05d3773ea6e82137836637d00b9b89859471e-electron-linux

[pascalandr]

@CodeNomadBot

I want you to perform the gatekeeper checks for latest PR state

check for regressions
check for better possible implementations
Be the PR gatekeeper
Be ruthless about code quality

[codenomadbot]

Gatekeeper review complete for latest PR state.

Request understood: perform a strict review for regressions, implementation quality, and whether this should pass as PR gatekeeper. Assumption: review-only, no code changes requested.

Findings: none blocking.

Review notes:

• The fix correctly mirrors the permission queue pattern for question entries: duplicate request IDs are de-duped from byMessage, the queue entry is replaced in-place, the queue is sorted by preserved enqueue time, and active is recalculated from the first queue item.
• Preserving enqueuedAt with Math.min(existing.enqueuedAt, entry.enqueuedAt) addresses the reported deadlock/order regression where a global question later resolves to a tool part with a newer timestamp.
• Keeping newer messageId/partId while preserving original ordering is the right behavior for global-to-tool resolution.
• Regression coverage is targeted and meaningful: duplicate global-to-tool movement, active recalculation after removal, and original enqueue ordering across multiple questions.

Validation run:

• npm run typecheck --workspace @codenomad/ui passed.
• node --import tsx --test packages/ui/src/stores/message-v2/instance-store.test.ts passed, 5 tests.

Environment note:

• The worktree initially had no installed root dependencies, so the first validation attempt failed to resolve packages like solid-js and tsx. I ran npm ci from the existing lockfile and reran validation successfully.
• The worktree has an unrelated local modification in .opencode/package-lock.json; I ignored it for this PR review.

Non-blocking quality note:

• packages/ui/src/stores/message-v2/instance-store.ts is already large at about 1,288 lines. This PR only adds a small focused block, so I would not block or force a refactor here, but future message-store work should consider splitting queue/interrupt state helpers out of the monolith.

Gatekeeper verdict: approve from code-quality and regression-risk perspective.

--
Yours,
CodeNomadBot

[pascalandr]

I’m not sure this fully fixes #448.
This PR fixes a narrow question upsert/order case, but the issue discussion points to a broader problem: the blocking interruption is not visible/actionable, and the user gets stuck on “Waiting for earlier responses”.
This PR does not touch activeInterruption, the permission/question center, auto-opening behavior, or any way to act on the blocking item.
Reproduction validation from the reporter is needed.