fix(tools): use /images/generations endpoint for Gemini and OpenAI

[xthanhn91]

Summary

• create_image exclusively used /chat/completions with modalities:["image","text"], which only works on OpenRouter
• Gemini returns HTTP 400: "Image generation is not yet supported on the chat.completions endpoint for this model"
• OpenAI DALL-E models also require /images/generations, not /chat/completions
• Route OpenRouter → /chat/completions (supports modalities), all other providers → /images/generations
• Update default Gemini model from deprecated gemini-2.0-flash-exp to gemini-2.5-flash-image

Root Cause

The callImageGenAPI function was designed for OpenRouter's modalities-based chat completions format and used for all providers. But only OpenRouter supports this — Gemini and OpenAI both require the standard /images/generations endpoint.

Gemini error response:

{
  "error": {
    "code": 400,
    "message": "Image generation is not yet supported on the chat.completions endpoint for this model. Please use the standard client.images.generate method for creation"
  }
}

Fix

Added callStandardImageGenAPI using the /images/generations endpoint with response_format:"b64_json" — the standard OpenAI-compatible image generation format supported by Gemini, OpenAI, and most other providers.

Provider	Endpoint	Status
OpenRouter	/chat/completions + modalities	✅ (unchanged)
Gemini	/images/generations + b64_json	✅ (fixed)
OpenAI	/images/generations + b64_json	✅ (fixed)

Test plan

• Test with Gemini provider: create_image with gemini-2.5-flash-image
• Test with OpenRouter provider: verify modalities-based flow still works
• Test with OpenAI provider (if available): verify DALL-E image generation
• Verify MEDIA: path returned and image delivered to channel