Improve Lemonade instructions by adding missing commands, improving prompts, and clarifying model usage

[adamlam2-amd]

• updating links
• updating models to use. Re-use the Gemma one and use a newer qwen3.5-35b-a3b model. This clarifies the model usage and stops introducing new models (like Qwen3.5-4B) unnecessarily.
• clearer prompt instructions
• removed unnecessary curl commands
• added necessary venv commands
• put sample output in another file which can be displayed to the user

[adamlam2-amd]

@sreeram-11 can you fix the tests please

[danielholanda]

@adamlam2-amd Can you be more specific when creating PR titles? Something like Standardize model usage and clean up Lemonade Playbook could be good here.

[ldokovic-personal]

did we separate the ports distinction here and in flashcards.py(R7)? Lemonade is running at :8000 by default on AI Halo

[adamlam2-amd]

lemonade should be at 13305 by default with recent versions of Lemonade. Let me know if you see differently.

[sreeram-11]

@sreeram-11 can you fix the tests please

Fixed!

FYI, tests were failing because the Gemma-4-E2B-it-GGUF model was not downloaded in those runners.