Add Qwen 3.5 GGUF support and configurable chat-template kwargs (thinking toggle)

[Copilot]

Qwen 3.5 GGUF models were not loading reliably, and there was no way to pass llama.cpp chat-template kwargs needed to disable default thinking mode.
This change adds first-class support for chat-template kwargs in GGUF config flow and applies sensible defaults for Qwen 3.5 models.

• GGUF inference schema: add chat-template kwargs

  • Added chatTemplateKwargs to GgufInferenceParams.
  • Preserved JSON compatibility by keeping it optional/empty by default.

• Engine wiring: pass kwargs to native runtime

  • GGUFEngine now applies chat-template kwargs after model load.
  • Uses reflective invocation of nativeSetChatTemplateKwargs(String) for backward compatibility with older native libs that do not expose this API.

• Qwen 3.5 default behavior

  • Added GgufEngineSchema.defaultsForModel(modelName) to preconfigure Qwen 3.5 models with:
    • {"enable_thinking": false}
  • Used this default schema in both model install paths:
    • local import/install flow
    • downloaded model install flow

• Config UI support

  • Added an editable GGUF config field:
    • Chat Template Kwargs (Optional JSON)
  • Added corresponding ViewModel update path so users can override defaults per model.

• Model catalog update

  • Added default repository entry for:
    • Qwen/Qwen3.5-4B-Instruct-GGUF

Example of newly supported config payload:

{
  "chatTemplateKwargs": "{\"enable_thinking\": false}"
}

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.