Validate the generalized design against existing Qwen support

[michalharakal]

Summary

Use the existing Qwen support as a validation target for the generalized architecture.

Scope

• Verify the generalized provider/capability design works with the existing Qwen implementation
• Ensure no Qwen-specific logic leaks into unrelated shared code
• Confirm end-to-end tool-calling still works when resolved through the new architecture

Suggested scenario

1. Load a Qwen instruct model

2. Run ToolCallingDemo through the generalized resolution path

3. Ask a simple tool-use question such as listing files

4. Verify:

   • provider selection works correctly
   • tool calls are parsed and executed correctly
   • the agent loop completes successfully

Acceptance criteria

• Existing Qwen support works unchanged or with minimal integration changes
• Qwen validates the generalized architecture successfully

---

Initial implementation notes

Existing relevant areas

• llm-agent
• llm-inference/qwen
• ToolCallingDemo.kt
• existing parser/template test suites

Suggested placement

• shared interfaces and generic logic: llm-agent
• model-specific implementations: dedicated existing model modules