Refactor backward explanations into shared structured model

Implement min reduction function with equal-share subgradient at ties

Refine backward semantic visual language with plum gradient contributions and strict adjoint invariants

Implement stepwise text trace, semantic diagram model, and quiet cluster styling

Formally revise visual grammar with adjoints, two-part trapezoids, and Sigma accumulation

Add Mathbook presentation option for LaTeX partial derivative notation

Implement PyTorch conventions, vector indexing, class operations, and layout positioning

Add compact vector and matrix support with adjoint backpropagation