Provide a default LLM well tuned for CQL and CQL Studio

[preston]

The coder models we've been using are not performing well for CQL and we likely need something tuned for this purpose.

[preston]

We are going we several of the qwen3-based variants, specifically:

• qwen3-coder-next:latest
• qwen3-coder:latest
• qwen3.5:cloud (for users that can't run models locally)