feat(conversation): add Auto planner/worker foundation for aionrs

[JeffC0628]

Summary

• Add product-level Auto for aionrs: planner / worker / utility slot settings, sticky Auto in selectors (desktop + mobile + Guid), and conversation.extra.auto_model.
• Phase 1 resolves a concrete worker-slot model into conversation.model (never sends the Auto sentinel to Core). Full long-run cost savings need Phase 2 (switch model without agent rebuild).
• Includes PRD, i18n (13 locales), and unit tests for resolve heuristics.

Closes #4143

Scope / non-goals

• Does not change assistant mode: auto (Remember last used).
• Does not force Auto as the default model.
• ACP platforms unchanged.

Test plan

• Configure Auto slots under Settings → Model; verify persistence across restart
• In aionrs chat, pick Auto → pill shows Auto · {model}; extra.auto_model.enabled true
• Pick a concrete model → Auto flag clears
• Guid new chat with Auto → create stores auto_model on the conversation
• Mobile action sheet shows Auto and selects correctly
• bun run test -- tests/unit/renderer/autoModel/resolveAutoModel.test.ts
• bun run i18n:types + node scripts/check-i18n.js

[codecov]

Codecov Report

❌ Patch coverage is 36.93694% with 140 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...s/desktop/src/renderer/utils/autoModel/settings.ts	20.00%	20 Missing ⚠️
...p/src/renderer/hooks/agent/useAutoModelSettings.ts	0.00%	17 Missing ⚠️
...renderer/pages/guid/hooks/useGuidModelSelection.ts	0.00%	14 Missing ⚠️
...p/src/renderer/utils/autoModel/resolveAutoModel.ts	70.21%	4 Missing and 10 partials ⚠️
...top/src/renderer/utils/autoModel/tierHeuristics.ts	53.84%	6 Missing and 6 partials ⚠️
...c/renderer/pages/guid/components/GuidActionRow.tsx	8.33%	11 Missing ⚠️
...tings/SettingsModal/contents/ModelModalContent.tsx	52.38%	7 Missing and 3 partials ⚠️
...versation/platforms/aionrs/AionrsModelSelector.tsx	16.66%	4 Missing and 6 partials ⚠️
...sation/platforms/aionrs/useAionrsModelSelection.ts	37.50%	10 Missing ⚠️
...es/conversation/platforms/aionrs/AionrsSendBox.tsx	10.00%	9 Missing ⚠️
... and 3 more

📢 Thoughts on this report? Let us know!

[IceyLiu]

Reviewed together with #4146 — the feat/auto-phase2-routing branch already contains all commits from this PR, so the full review is posted there: #4146 (comment)

Local gate on the Phase 2 branch is green (tsc 0 errors, 9/9 unit tests, i18n in sync across all 13 locales, lint 0 errors), and the Phase 1 foundation here looks clean.

Two items in that review land on Phase 1 code specifically, so flagging them here as well:

1. requireVision is never passed. handleSelectAuto in useAionrsModelSelection.ts calls resolveAutoModel without it, so the vision filter inside resolveAutoModel is dead code and Auto can resolve a text-only model for an image send.
2. The tier heuristics rank the wrong models and the cost/balance/quality preference has no observable effect. Over a realistic catalogue, gemini-3-pro-preview always wins planner (ahead of Opus and GPT-5, via double-scoring on /pro-preview/i + /gemini.*pro/i), qwen3-coder-plus always wins worker (double-scoring /qwen.*plus/i + /coder/i), and all three preferences select identical winners. tierHeuristics.ts:41 is also an identity expression — Math.max(utility, planner > 6 ? 0 : utility) === utility — so the intended utility zeroing never takes effect.

Also: the utility slot is configurable in Settings but not consumed by any resolve path yet, and AutoModelSettingsBlock renders above the "no configured models" empty state in ModelModalContent.tsx.

Suggest consolidating onto #4146 and closing this one once the merge order is settled.

[JeffC0628]

Superseded by #4146 (includes Phase 1 + Phase 2).