fix: prefer Opus 4.6 1M for workers, don't switch desktop session from mobile

[PureWeen]

Changes

1. Worker models prefer Claude Opus 4.6 1M

• All 3 built-in presets (PR Review Squad, Implement & Challenge, Skill Validator) now use claude-opus-4.6-1m for orchestrators
• Added ModelHelper.ResolvePreferredModel() — resolves preferred model against available models at runtime, falling back to claude-opus-4.6 if 1M isn't available
• Fallback is applied in CreateGroupFromPresetAsync for both orchestrator and worker sessions

2. Mobile session switching no longer changes desktop view

• Removed SetActiveSession() call from WsBridgeServer.SwitchSession handler
• Mobile can browse and send to any session without disrupting the desktop user's active view
• History is still sent to the requesting mobile client as before

Tests

• 3,118 tests passing, 0 failures
• Updated GroupPresetTests and MultiAgentScenarioTests for new model defaults
• Updated SwitchSession_BroadcastsUpdatedActiveSession to verify desktop session is NOT changed

[PureWeen]

🤖 Multi-Model Code Review — R1 (PR #487)

Reviewers: Claude Opus 4.6, Claude Sonnet 4.6, GPT-5.3-Codex
CI Status: ℹ️ No CI checks configured
Prior comments: None (fresh review)

---

Summary

Two changes: (1) Built-in presets now prefer claude-opus-4.6-1m with runtime fallback via ModelHelper.ResolvePreferredModel(), (2) Mobile session switching no longer changes the desktop's active session.

---

Consensus Findings

No consensus findings. Four observations were each flagged by only 1/3 models:

Observation	Model	Notes
Worker prompts lost model names → workers can't dispatch specific models	Opus (1/3)	Sonnet/Codex: Workers use the task tool with model parameter; model assignment happens at session creation via preset.WorkerModels[i], not from prompt text. The prompt change to generic descriptions ("deep reasoning", "fast pattern matching") doesn't affect task tool dispatch.
Test assertion ?? "switch-b" makes NotEqual brittle on null	Sonnet (1/3)	Opus/Codex: In the test setup, CreateSessionAsync("switch-a") sets active session first, so GetActiveSession() is never null. The assertion correctly verifies NotEqual("switch-b", "switch-a").
Remote mode sends unresolved model to server	Opus (1/3)	The remote path sends preset metadata to the server, which does its own model resolution. Not a desktop-side bug.
SetSessionPreferredModel stores unresolved model after fallback	Codex (1/3)	Opus/Sonnet: The preferred model metadata is intentional — it records user intent. EnsureSessionModelAsync re-resolves at runtime.

---

Architecture Assessment (3/3 agree)

• ResolvePreferredModel fallback logic: Correct. Null/empty availableModels → returns preferred (safe, SDK handles). Match found → returns it. No match → tries fallbacks → last resort returns preferred. ✅
• Mobile switch isolation: Clean removal of SetActiveSession(). BroadcastSessionsList() correctly sends unchanged desktop state. SendSessionHistoryToClient sends history to requesting client only. ✅
• Test updates: Tests properly verify the new behavior — desktop active session stays unchanged after mobile switch. ✅
• Prompt text changes: Model name references removed from worker prompts, replaced with generic role descriptions. Workers still receive model assignment via WorkerModels[i] at session creation. ✅

---

Test Coverage

• ✅ SessionOrganizationTests updated for new model defaults (claude-opus-4.6-1m)
• ✅ WsBridgeIntegrationTests updated to verify desktop session NOT changed on mobile switch
• 🟢 ResolvePreferredModel has no dedicated unit test — logic is simple (3 branches) and tested indirectly via preset creation. Non-blocking.

---

Verdict: ✅ Approve

Clean PR. Model fallback logic is correct, mobile switch isolation is properly implemented, tests are updated. No consensus issues found across 3 models.

[PureWeen]

Split into separate PRs: #498 (mobile UI fixes) and #499 (model preferences + prompt changes)