Add native support for converting Qwen2 family models (including Qwen2.5) from safetensors to gguf format so we can run it.