Add Google VertexAI Provider Support

This PR adds support for Google VertexAI models (Gemini) as a new provider in the Hub, allowing users to route their LLM requests to Google’s models through our unified API interface.

Definition of Done

• Implementation of VertexAI provider
  • Chat completions endpoint with Gemini models
  • Text completions endpoint
  • Embeddings endpoint with textembedding-gecko
  • Tool/function calling support
  • Streaming support
  • Multi-modal support
• Unit tests
  • Provider configuration tests
  • Request/response conversion tests
  • Error handling tests
  • Model format conversion tests
• Integration tests
  • Chat completions test with recorded responses
  • Completions test with recorded responses
  • Embeddings test with recorded responses
  • Tool calling test with recorded responses
  • Test cassettes for offline testing
  • Quota retry mechanism
• Documentation
  • Configuration examples in README
  • Authentication methods (API key and service account)
  • Model support documentation
  • Usage examples with OpenAI SDK
• Configuration
  • Example configuration in config-example.yaml
  • Support for both API key and service account auth
  • Required parameters (project_id)
  • Optional parameters (location, credentials_path)
• Error Handling
  • Proper status code mapping from Google API
  • Informative error messages
  • Quota limit handling with configurable retries
  • Authentication error handling
• Code Review and Approval

Changes Made

1. Added new VertexAI provider implementation

   • Full support for Gemini models (chat, completion)
   • Text embeddings with textembedding-gecko
   • Tool/function calling with proper mapping
   • Streaming support for real-time responses
   • Multi-modal capabilities for image+text inputs

2. Added comprehensive test suite

   • Unit tests for core functionality
   • Integration tests with recorded responses
   • Test cassettes for offline testing
   • Quota retry mechanism for rate limits
   • Both auth methods covered

3. Updated documentation

   • Detailed VertexAI setup instructions
   • Authentication configuration guide
   • Model compatibility list
   • Usage examples with OpenAI SDK
   • Configuration parameter reference

4. Added robust configuration support

   • API key authentication
   • Service account authentication (JSON key file)
   • Project ID configuration
   • Regional endpoint configuration
   • Default values for optional parameters

Testing

• Unit tests: cargo test
• Integration tests with credentials:

  # Service Account Auth
  export VERTEXAI_CREDENTIALS_PATH="../credentials/vertexai-key.json"
  
  # Record new responses
  RECORD_MODE=1 cargo test
  
  # Replay mode (default)
  cargo test
  

• Test retry configuration:

  export RETRY_DELAY=60  # Seconds between retries
  

Security Considerations

• Credentials stored outside repository
• Test cassettes cleaned of sensitive data
• Support for both API key and service account auth
• Environment variable support for credentials
• No hardcoded sensitive values

Notes

• Default location is us-central1 (configurable)
• Automatic retry on quota limits (configurable)
• Test cassettes provided for offline development
• Compatible with existing OpenAI SDK clients

Fixes #19 /claim #19