Add comprehensive documentation for dual-model (low-cost vs normal) usage

[Copilot]

This PR adds comprehensive documentation for Mini-A's dual-model feature that enables cost optimization by intelligently switching between a high-capability main model and a low-cost model.

What's New

Environment Variables

• Documents the optional OAF_LC_MODEL environment variable alongside the existing OAF_MODEL
• Provides clear setup examples for both single and dual-model configurations

Dual-Model System Documentation

The documentation now explains how Mini-A's intelligent model selection works:

• Step 0: Always uses the main model for critical initial planning
• Routine Operations: Uses the low-cost model for summarization and simple tasks
• Smart Escalation: Automatically switches to the main model when:
  • 2+ consecutive errors occur
  • 3+ consecutive thoughts without action
  • 5+ total thoughts (thinking loop detection)
  • 4+ steps without meaningful progress
  • Repeating similar thoughts are detected

Practical Examples

Added real-world configuration examples for:

• OpenAI models (GPT-4 + GPT-3.5-turbo)
• Mixed providers (OpenAI + Ollama)
• Google Gemini models (Pro + Flash)

Enhanced Troubleshooting

New troubleshooting section covers dual-model specific issues like:

• Configuration format validation
• Frequent escalation patterns
• Invalid JSON fallback behavior
• Cost optimization verification

Benefits for Users

Cost Savings: Users can significantly reduce API costs by using cheaper models for routine tasks while maintaining quality for complex operations.

Transparency: Clear logging shows which model is handling each operation, making it easy to monitor cost optimization.

Automatic Quality Assurance: The smart escalation system ensures complex scenarios always get the attention of the more capable model.

Files Changed

• README.md: Added dual-model configuration section and environment variable documentation
• USAGE.md: Comprehensive dual-model guide with examples, troubleshooting, and best practices

The documentation is designed to be user-friendly with practical examples that users can copy and adapt for their specific needs.

Original prompt

Update documentation with the low-cost vs normal model usage

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.