Changes in version 0.99.0                        

  - Added public PLS, PCA, SVD, prediction, evaluation and
    cross-validation interfaces for CPU, optional CUDA and optional
    Apple Metal backends.
  - Added optional classification heads for PLS-DA, including argmax,
    latent-space LDA and candidate-kNN.
  - Added package datasets, examples, benchmark scripts and a single
    user vignette.
  - CUDA and Metal builds are optional; CPU-only installation remains
    the default.