Unsloth updates Mistral Small 4 quants

This is less a flashy launch than a practical correction: when a big local model behaves badly, the quant/template layer is often the real bug.

• –The repo explicitly calls out “Unsloth chat template fixes,” which is the kind of downstream fix that can materially change output quality
• –For local runners, updated GGUFs matter more than benchmark slides because broken prompting can make a strong base model look mediocre
• –Mistral Small 4 is still a heavyweight 119B MoE model, so these quants are about making it usable, not making it cheap
• –The note to use `--jinja` in llama.cpp suggests the release is also about interoperability, not just compression
• –If you compared older quants and got mixed results, this is the version worth re-testing before writing the model off