feat: add Gemini API support to OpenAI provider

[douglaz]

Summary

Adds support for Google Gemini models through the OpenAI-compatible API endpoint, enabling the use of advanced models like Gemini 2.5 Flash and 2.5 Pro.

Changes

• Added support for GEMINI_API_KEY environment variable as a fallback option
• Automatically fixes Gemini API URLs to include the required /chat/completions endpoint
• Updated error messages to reflect support for both OpenAI and Gemini APIs

Technical Details

The Gemini API provides an OpenAI-compatible endpoint at https://generativelanguage.googleapis.com/v1beta/openai/. This change detects when this URL is used and automatically appends /chat/completions if needed, ensuring compatibility with the existing OpenAI provider implementation.

Testing

• All existing tests pass
• Manually tested with Gemini 2.5 Flash model in the Fedimint challenge
• Agent successfully connects and uses Gemini for reasoning

Benefits

• Access to Gemini's latest models including 2.5 Flash (fast, efficient) and 2.5 Pro (advanced reasoning)
• No need for a separate Gemini provider - reuses existing OpenAI infrastructure
• Seamless switching between OpenAI and Gemini models via configuration

Configuration Example

[llm]
provider = "openai"
model = "gemini-2.5-flash"
api_url = "https://generativelanguage.googleapis.com/v1beta/openai"
api_key = "YOUR_GEMINI_API_KEY"
temperature = 0.3
max_tokens = 8000