Algorithmic and dynamical stability

This entrypoint collects the stability chapter. We keep three pieces together because they are often used in the same arguments, while still leaving each piece in its own source file:

• Stability.Core defines datasets, replace-one/remove-one perturbations, learning maps, losses, population error, empirical error, and standard algorithmic-stability notions.
• Stability.Dynamics gives the discrete-time dynamical-system vocabulary used by recurrent models, samplers, and stability diagnostics.
• Stability.RidgeRegression1D is the worked theorem development: a concrete, fully proved stability analysis for bounded one-dimensional ridge regression.

References:

• Bousquet and Elisseeff, "Stability and Generalization", JMLR 2002.
• Shalev-Shwartz et al., "Learnability, Stability and Uniform Convergence", JMLR 2010.
• Hardt, Recht, and Singer, "Train faster, generalize better: Stability of stochastic gradient descent", ICML 2016.