fix(chat): preserve queued follow-ups while the active head loads

[mameikagou]

What does this PR do?

Preserve accepted chat follow-ups when their queued acknowledgements arrive before the client has loaded the active pending head. Previously, every acknowledgement replaced the queue-only cache with [task], dropping earlier queued entries from local pending state.

Thinking path:

• Chat send acknowledgements and the authoritative pending query settle asynchronously.
• A response with queued: true can arrive before the running head is available locally. The shared pending helper deliberately keeps such state headless until the pending query loads.
• More than one acknowledgement can arrive in this window. Replacing queued_tasks loses earlier accepted entries; merging through the existing normalizeQueue preserves them with the same FIFO ordering, task-ID deduplication, and rich-preview preference as the existing loaded-head path.
• Both the main chat and floating chat window use this helper, so the fix belongs in this shared state layer, with regression tests beside it.

Scope: this fixes a deterministic frontend queue-state overwrite. It does not change backend scheduling, cancel/retry behavior, or claim that persisted messages were deleted. It has not established that every reported instance of a replaced message has this cause.

Related Issue

User-reported repeated chat queue replacement; no linked GitHub issue.

Type of Change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Refactor / code improvement (no behavior change)
• Documentation update
• Tests (adding or improving test coverage)
• CI / infrastructure

Changes Made

• packages/core/chat/pending.ts: append queued acknowledgements to the existing headless queue and normalize, rather than replacing it. Preserve supports_queue and keep the active head unspecified until it is known.
• packages/core/chat/pending.test.ts: four regression cases covering sequential and reversed acknowledgements, late head acknowledgement, task-ID deduplication, sibling preservation, and rich/sparse preview observations in either order. Also declare the DOM-free Node test environment.

How to Test

1. Begin without a cached active head. Accept follow-up A with queued: true, then accept follow-up B with queued: true. Before this fix the cached queue changes from [A] to [B]; after it the queue is [A, B] with no fabricated head. Reversing acknowledgement order produces the same FIFO queue.
2. Confirm that a later head acknowledgement preserves both follow-ups. Re-observe one queued task with a sparse or rich payload and verify the sibling and message preview remain intact.
3. Run the regression suite and checks below from the repository root:

pnpm --filter @multica/core exec vitest run chat realtime/use-realtime-sync.test.ts api/schemas.test.ts --maxWorkers=2
pnpm --filter @multica/core typecheck
pnpm --filter @multica/core exec eslint chat/pending.ts chat/pending.test.ts
git diff --check

Verified locally: 10 test files / 320 tests pass; core typecheck, changed-file lint, and diff whitespace check pass. The four new regression cases all failed against the original helper before applying the fix. Dependencies were reused from an existing local installation; no dependency or lockfile changes are included.

Risks / verification limits: this reuses existing normalization semantics and only changes the headless queued: true branch. It was manually reviewed, but no browser E2E, real-provider smoke test, full-monorepo build, or deployment was performed. No server or daemon changes are included.

Checklist

• I have included a thinking path that traces from project context to this change
• I have run tests locally and they pass
• I have added or updated tests where applicable
• If this change affects the UI, I have included before/after screenshots (no browser screenshots captured; deterministic state reproduction above)
• I have updated relevant documentation to reflect my changes (not applicable: internal state fix, no public API changes)
• If I added a new runtime / coding tool / UI tab, I synced the change to landing copy (apps/web/features/landing/i18n/) and relevant docs (apps/docs/content/docs/) (not applicable)
• If this PR touches Chinese product copy, I checked it against apps/docs/content/docs/developers/conventions.zh.mdx (terminology, mixed-rule for task / issue / skill) (not applicable)
• I have considered and documented any risks above
• I will address all reviewer comments before requesting merge

AI Disclosure

AI tool used: Codex via Multica.

Prompt / approach: The user reported that sending additional chat messages before a response replaced earlier messages instead of queueing them, and requested an upstream PR. Inspected the send and pending-state paths, reproduced the shared helper's headless queue overwrite, added failing regression tests, reused existing queue normalization for a minimal fix, ran related core tests/typecheck/lint, and manually reviewed the two-file diff. No live chat messages or provider requests were used to test this patch.

Screenshots (optional)

No visual styling changes or browser screenshots. The reproducible queue-state trace is [A] -> [B] before and [A] -> [A, B] after the fix.

[vercel]

@mameikagou is attempting to deploy a commit to the IndexLabs Team on Vercel.

A member of the Team first needs to authorize it.