SVM as an NNModuleSpec

The SVM spec model includes a gradient-descent baseline and prediction helpers. This file adds the NNModuleSpec wrapper so it can be composed/exported in the module system.

References:

• For the actual SVM objective/gradients and classic SVM citations (Cortes-Vapnik, Vapnik), see NN.Spec.Models.Svm: NN/Spec/Models/Svm.lean.
• PyTorch analogy for the forward map: a linear score X @ w + b is the same shape-level role as torch.nn.Linear(p, 1) (no activation).

def Spec.svmModule {α : Type} [Context α] {p : ℕ} (model : SVM p 0 α) : ModSpec.NNModuleSpec α (Shape.dim p Shape.scalar) (Shape.dim 1 Shape.scalar)

Linear SVM wrapped as an NNModuleSpec (single-output linear layer).

def Spec.svmChain {α : Type} [Context α] {p : ℕ} (model : SVM p 0 α) : ModSpec.SpecChain α (Shape.dim p Shape.scalar) (Shape.dim 1 Shape.scalar)

SpecChain wrapper for linear SVM.