PCA and kNN for Bigarrays

• admonition.warning

  This library was vibe-coded with AI assistance and has not been thoroughly reviewed or tested. Use at your own risk and expect breaking changes.

PCA dimensionality reduction and k-nearest-neighbours classification for float32 Bigarray data. Pure OCaml with no external dependencies.

Used by the tessera pipeline to reduce high-dimensional satellite embeddings to 2–3 components for visualization, and to classify pixels against labelled training data.