feat: update default agent to use first_available model selection

[dgageot]

The default agent now uses the new first_available model selection mechanism introduced in the previous commit. Instead of relying on environment detection at startup, the agent tries a prioritized list of models in order: Claude Sonnet 4.6, GPT-5, Gemini 3.5 Flash, Mistral Small, and Qwen3 (local fallback).

This change also bumps the default model versions for cloud providers. OpenAI moves from gpt-5-mini to gpt-5, and Google moves from gemini-2.5-flash to gemini-3.5-flash. The test expectations are updated to reflect these newer versions.

No breaking changes; the agent continues to work seamlessly across different environments without requiring users to reconfigure anything.

[docker-agent]

Assessment: 🟡 NEEDS ATTENTION

[docker-agent]

[LIKELY HIGH] Potentially non-existent Google model name gemini-3.5-flash

The DefaultModels map was updated from "gemini-2.5-flash" to "gemini-3.5-flash", and the same name appears in default.yaml (google/gemini-3.5-flash).

Google's Gemini versioning has followed the series 1.0, 1.5, 2.0, 2.5 — a jump to 3.5 is a significant version skip. As of 2026-06-03, gemini-3.5-flash does not appear in Google's Gemini API catalog, while gemini-2.5-flash is a confirmed, released model.

Impact: Any user whose sole configured key is GOOGLE_API_KEY (or GOOGLE_GENAI_API_KEY) will receive an API error ("unknown model" or similar) when the first_available resolver or DefaultModels auto-detection selects the Google provider. The two affected code paths are:

1. default.yaml line 6: - google/gemini-3.5-flash (new first_available path)
2. auto.go line 58: "google": "gemini-3.5-flash" (legacy DefaultModels path)

If the intent is to bump to a newly available model, please verify the exact model identifier in the Google Generative AI model docs before merging. If it should remain the previous version, revert to "gemini-2.5-flash".

[docker-agent]

[LIKELY HIGH] Potentially non-existent Google model name gemini-3.5-flash

Same issue as flagged in auto.go — this YAML entry uses the same model name google/gemini-3.5-flash. If gemini-3.5-flash is not a valid Google API model identifier, users with only Google credentials configured will fall through to dmr/ai/qwen3 (local fallback) unexpectedly, or fail with an API error if google is the resolved provider.