Changes in version 0.2.0                        

  - New R core solver: soft_sinkhorn() (Anderson-accelerated
    semi-relaxed soft Sinkhorn), row_conditional(), make_cost(),
    row_ce_loss(), zscore_phi().
  - New fitting: fit_iot() with exact implicit gradients, two-stage
    debias refit and multi-restart; print/summary methods for
    bioIOT_fit.
  - New trajectory layer: transition_matrix(),
    pseudotime_from_transition(), build_state_features().
  - New single-cell interfaces: runIOT() generic with matrix,
    SingleCellExperiment and Seurat methods.
  - New visualisation (ggplot2): plot_transition_heatmap(),
    plot_transition_flow(), plot_theta(), plot_pathway_trend().
  - New reproducible demo: simulate_iot_states() plus bundled dataset
    demo_iot_states.
  - New vignette and expanded test suite (implicit-gradient vs finite
    differences, recovery, interfaces, plots).

                        Changes in version 0.1.0                        

  - Initial release: pathway_markers, score_pathways, collapse_probes,
    gsm_id, has_cel_file, find_platform_file.