Module mistral 

Mistral Model V2 - Clean implementation using solid abstractions

This implements the Mistral architecture which is used by:

• Mistral-7B, Mistral-7B-Instruct
• Key features: Sliding window attention, GQA
• Uses unified Tensor type from tensor_core
• Implements Model trait from model_core
• Supports loading via weight_loader_core

Structs

MistralAttention

Mistral attention mechanism with sliding window

MistralConfig

MistralLayer

Mistral transformer layer

MistralMLP

Mistral MLP (feed-forward network)

MistralModelV2

Main Mistral model implementation