docs: fix Gemma 4 Colab notebook

[abetlen]

Summary

• Move the Gemma 4 Colab badge out of the multimodal model table row so it renders cleanly in the README.
• Switch the notebook default to the Q8_0 Gemma 4 weights.
• Replace the text-only smoke prompt with a normal question.
• Fix the multimodal image URL by pointing at the upstream llama.cpp raw image that currently resolves.

Validation

• Validated the notebook JSON structure and confirmed it has no saved outputs or execution counts.
• Verified the replacement image URL returns HTTP 200.