[codex] Refresh default LLM provider models

[warren618]

Summary

• Update provider defaults for OpenRouter, OpenAI, DeepSeek, Gemini, Groq, DashScope/Qwen, Zhipu, Moonshot/Kimi, and MiniMax per [Feature] Update LLM providers to latest models (2026) #133.
• Sync agent/.env.example, CLI init provider choices, interactive onboarding suggestions, SKILL metadata, and README recommended-model examples.
• Add regression coverage so the provider registry, legacy CLI init choices, and onboard defaults do not drift back to stale model IDs.

Why

#133 reports that several bundled LLM provider defaults have fallen behind the currently available model families. Users who rely on generated setup files or the init/onboarding flow should start from the newer model IDs instead of deprecated or older defaults.

Validation

• cd agent && python -m json.tool src/providers/llm_providers.json >/dev/null
• cd agent && python -m py_compile cli/_legacy.py cli/onboard.py cli/intro.py src/providers/llm.py
• cd agent && python -m pytest tests/test_llm_provider_defaults.py tests/test_cli_init.py tests/test_settings_api.py tests/test_openai_codex.py tests/test_kimi_reasoning_content.py -q -> 48 passed, 2 warnings
• git diff --check

Closes #133