feat: add native Ollama API support with "think" control

[jordibeen]

While experimenting with Ollama's thinking control feature using "think": false, I discovered this parameter wasn't being respected. This is because this plugin currently uses Ollama's OpenAI-compatible endpoint (/api/v1/chat/completions) rather than the native API.

According to Ollama's documentation, their OpenAI compatibility is "experimental and subject to major adjustments including breaking changes," while recommending their native API for "fully-featured access." This limitation prevents access to Ollama-specific features like thinking control.

This PR implements native Ollama API support to unlock these features.

Changes Made

• Native API Response Parsing - Added Ollama-specific streaming response handler for {"message":{"content":"text"}} format with robust error handling
• Thinking Parameter Support - Implemented think parameter control for models like Qwen, with example ChatQwen3-8B agent configuration
• Proper Parameter Structure - Refactored payload to follow Ollama API specification with parameters nested under options object (temperature, top_p, min_p, num_ctx, top_k)
• Authentication Handling - Removed unnecessary auth headers for Ollama's local API