Lemonade 10.1 sharpens AMD local AI

This is less about a flashy new capability than about making local AI feel like a real daily driver on AMD hardware. The direction is clear: Lemonade is trying to become the control plane for local inference, not just another model runner.

• –The new `lemonade` CLI and simplified config flow lower the friction for first-time setup and repeat use
• –`lemonade launch codex|claude` is the most strategically interesting piece, because it pushes Lemonade into agent workflows rather than just chat inference
• –Gemma 4 GPU support broadens the model menu, which matters more than raw benchmark gains for local users who want choice
• –Super-resolution for generated images shows the project is moving beyond text-only serving into a broader multimodal local stack
• –The breaking changes are real: the default port changed, and `lemonade-server` is being deprecated, so this is a version to upgrade deliberately