AE1SVM: `nu` is documented as "Parameter for the SVM loss" but is never used

AE1SVM.__init__ accepts and documents nu:

# pyod/models/ae1svm.py:224
nu : float, optional (default=0.1)
    Parameter for the SVM loss.

It is stored at ae1svm.py:257 (self.nu = nu) and that is its only non-signature occurrence in the module. The training loss at ae1svm.py:334-336 is

svm_loss = torch.mean(torch.clamp(1 - svm_scores, min=0))
loss = self.alpha * recon_loss + svm_loss

a plain hinge with no ν term. AE1SVM(nu=0.01) and AE1SVM(nu=0.5) train identically.

Why it matters: ν is the defining hyperparameter of a one-class SVM (it bounds the fraction of training points treated as outliers and the fraction of support vectors); in the AE-1SVM paper (Nguyen & Vien, 2018) the OC-SVM term is (1/ν)·mean(max(0, ρ − w·φ(x))) − ρ. A user tuning nu gets no effect and no warning. Same shape as #714 (DevNet).

Proposed fix: either (a) wire nu into the loss with a learnable ρ per the paper, with a regression test that two nu values produce different decision_scores_; or (b) if the simplified hinge is intentional, remove nu from the signature/docstring with a deprecation. Happy to do either once you say which.

Found with a script that checks every detector for the sklearn parameter contract (get_params/clone/refit); drafted with Claude Code assistance and verified by hand.