Summary

• add first-class Mistral model family support via BaseModel
• wire engines/models registry for base, LoRA, INT8, LoRA+INT8, and LoRA+Kbit variants
• configure default generation and fine-tuning profiles for all mistral keys
• update supported model docs and README model tables
• add coverage tests for registry, config keys, and engine/model naming

Model source
Uses mistralai/Ministral-3-14B-Instruct-2512 from Hugging Face instead of Mistral-7B.

Issue
Closes #273