DataSplits.jl: Data splitting strategies for machine learning in Julia

DataSplits v0.3.1 Diff since v0.3.0 Fixed MDKSSplit and LazyMDKSSplit now build the default Mahalanobis metric from the inverse covariance matrix, as Distances.Mahalanobis expects. Sphere exclusion seeds clusters in index order, so cluster assignment is reproducible across Julia versions for a fixed RNG. Type instabilities in LazyDuplexSplit and LazySPXYSplit. Changed Faster Kennard–Stone, OptiSim, Duplex, FieldStrength, SPXY, Spectral and CombinatorialPurgedKFold implementations, with lower memory use. OptiSim keeps its selection order, so selected indices for a given seed can differ from 0.3.0. Rewritten documentation: introduction, getting started walkthrough, and the distance-based, cross-validation, time-series and group-aware pages. CI runs on Windows and macOS; pull requests report benchmark changes. Added contributing guidelines and complete citation metadata (CITATION.cff). Merged pull requests: [AUTO] pre-commit update (#73) (@github-actions[bot]) Benchmark action (#74) (@davide-grheco) Optimize optisim performance (#75) (@davide-grheco) Optimize Kennard Stone (#76) (@davide-grheco) Optimize Duplex algorithm (#77) (@davide-grheco) Optimize FieldStrengthSplit: triangle symmetry and @simd column updates (#78) (@davide-grheco) [AUTO] pre-commit update (#79) (@github-actions[bot]) Improve the documentation with reproducible examples (#80) (@davide-grheco) Joss ci hardening (#81) (@davide-grheco) Paper changes (#82) (@davide-grheco) Show a plot with performance change in PRs (#83) (@davide-grheco) Use partial Eigen decomposition in Spectral Cluster (#84) (@davide-grheco) Fix type instability in LazySPXY (#85) (@davide-grheco) Fix type instability in LazyDuplexSplit (#86) (@davide-grheco) Replace setdiff with functions which do not need to materialize a set (#87) (@davide-grheco) Use plain loops for argmax/argmin on float vectors (#88) (@davide-grheco) Use the inverse covariance for MDKS Mahalanobis distance (#89) (@davide-grheco) Fix mdks mahalanobis (#90) (@davide-grheco) Seed sphere exclusion clusters in index order (#91) (@davide-grheco) Return OptiSim selection as a vector (#92) (@davide-grheco) Normalise the SPXY distance matrices in place (#93) (@davide-grheco) Spectral inplace (#94) (@davide-grheco) Benchmark DataSplits against astartes (#96) (@davide-grheco) Add astartes comparison table (#97) (@davide-grheco) Shorten the manuscript (#98) (@davide-grheco)