Add Venice.ai as provider

[faces-of-eth]

Please explain the motivation behind the feature request

We need to add Venice.ai as a provider to enhance our AI capabilities while maintaining strong privacy guarantees and transparency. Venice.ai offers several unique advantages:

1. Privacy-First: Unlike many AI providers, Venice.ai is built with privacy as a core principle and does not store or utilize user data.
2. Open Source Foundation: All models are open-source, providing full transparency and auditability.
3. Diverse Model Selection: Access to powerful models like Llama 3.3 (70B), Mistral 3.1 (24B), and specialized models for coding (Qwen 2.5 Coder) and vision tasks (Qwen 2.5 VL).
4. OpenAI API Compatibility: Easy integration with existing OpenAI-based implementations.

This addition would give users more control over their data while maintaining high-quality AI capabilities.

Venice is also Agent-first, which is a good fit for Goose.

Autonomous AI Agents can programmatically access Venice.ai’s APIs without any human interaction using the “api_keys” endpoint. AI Agents are now able to manage their own wallets on the BASE blockchain, allowing them to programmatically acquire and stake VVV token to earn a daily VCU inference allocation. Venice’s new API endpoint allows them to automate further by generating their own API key.

Describe the solution you'd like

Implement Venice.ai as a provider with the following features:

1. Provider Configuration:

interface VeniceConfig {
  apiKey: string;
  baseUrl: string; // defaults to Venice's API endpoint
  defaultModel?: string; // defaults to "llama-3.3-70b"
}

2. Model Support Matrix:

• General Purpose: llama-3.3-70b (65K context), llama-3.1-405b (65K context)
• Code Optimization: qwen-2.5-coder-32b (32K context)
• Vision Capabilities: qwen-2.5-vl (32K context)
• Fast Inference: llama-3.2-3b (131K context)
• Reasoning: deepseek-r1-671b (131K context)

3. Feature Support:

• Function Calling
• Response Schema Support
• Web Search Integration
• Vision Processing
• Extended Context Windows (up to 131K tokens)

4. Integration with existing authentication and configuration systems

Describe alternatives you've considered

1. Using individual open-source models directly:

• Pros: Maximum control and customization
• Cons: Higher maintenance overhead, need for own infrastructure

2. Other privacy-focused providers:

• Pros: Similar privacy guarantees
• Cons: Smaller model selection, less specialized models

3. Self-hosting open-source models:

• Pros: Complete control over data and models
• Cons: Significant infrastructure costs, maintenance complexity

Additional context

Key Technical Details:

• All models support temperature and top_p adjustments
• Context windows range from 32K to 131K tokens
• Specialized models for different use cases (coding, vision, reasoning)
• OpenAI-compatible API makes integration straightforward

Implementation Priority:

1. Core API integration with general-purpose models
2. Specialized model support (code, vision)
3. Advanced features (function calling, web search)

Model Capabilities Overview:

Most Powerful: llama-3.1-405b (405B parameters)
Most Versatile: mistral-31-24b (vision, function calling, web search)
Fastest: llama-3.2-3b
Best for Code: qwen-2.5-coder-32b
Vision Support: qwen-2.5-vl

• I have verified this does not duplicate an existing feature request