Enhances provider compatibility by adding reasoning (thinking) content extraction and streaming support for:

• Anthropic models with extended thinking blocks
• Gemini models with thought parts
• Mistral models with Magistral array content format
• Ollama models with improved token reporting via stream_options
• OpenAI-compatible providers with thinking field handling

Key improvements:

• Streaming responses now yield separate reasoning chunks for better agent interaction
• All providers properly parse and expose reasoning content from model outputs
• Ollama stream_options now includes usage reporting for better token tracking