feat(agent): reuse chat Quick/Powerful model selector

[AnthonyRonning]

Summary

Agent Mode's composer now uses the same suggested Quick/Powerful model selector as Maple chat, instead of the Agent-only GLM 5.2 + Quick pairing.

OpenSecret already routes auto:quick and auto:powerful to the current recommended models (DeepSeek V4 Flash and GLM 5.3). Agent Mode no longer needs a separate recommended set.

Behavior

• The Agent composer reuses chat's ModelSelector (Quick / Powerful / More models), with Agent-owned selection, disable, and catalog loading.
• New Agent tasks default to Quick (auto:quick).
• The previous Agent-only GLM 5.2 default migrates to Quick in native config and sticky selectedAgentModel preferences.
• Explicit other model choices, including Powerful, stay as-is. Started tasks remain locked to their persisted model.

Testing

• bun --no-env-file test ./src/services/agentModels.test.ts
• Frontend format, lint, typecheck, and the full Bun suite (811 tests)
• Focused Rust default/migration tests, plus the pre-commit locked Rust suite
• Desktop GUI smoke was skipped per request

[cloudflare-workers-and-pages]

Deploying maple with    Cloudflare Pages

Latest commit:	807e43c
Status:	✅  Deploy successful!
Preview URL:	https://a75f8c72.maple-ca8.pages.dev
Branch Preview URL:	https://codex-agent-chat-models-mapl.maple-ca8.pages.dev

View logs