M19: Scientific imaging — ML segmentation, 3D ops, histology, registration, time series (506 tests)

1. ML Segmentation (new ri-ml crate): multi-scale feature extraction (Gaussian/
   LoG/Hessian/Sobel/DoG at 5 scales), random forest classifier from scratch
   (decision trees with Gini impurity, bagging, random feature subsets),
   train_from_rois, predict probabilities, save/load model
2. 3D Operations (new ri-3d-ops crate): gaussian_blur_3d (separable),
   median_filter_3d, dilate_3d/erode_3d, distance_transform_3d,
   connected_components_3d (6/26 connectivity), measure_3d_objects
   (volume/surface/centroid/sphericity), resample_3d (trilinear)
3. Color Deconvolution: Beer-Lambert unmixing, presets (H&E, H-DAB,
   Masson Trichrome, Alcian Blue), auto stain vector estimation (PCA/NMF)
4. Advanced Registration: landmark-based (rigid/similarity/affine via
   least squares), B-spline non-rigid, Thirion's demons deformable
5. Time Series: bleach correction (ratio/exponential/histogram matching),
   temporal median/difference, delta_F/F0 for calcium imaging,
   reslice, orthogonal views
6. OME-TIFF: XML metadata parser, multi-series support, tiled TIFF,
   bio_formats_metadata common struct (pixel_size, z_step, channels)
7. Frequency Domain: notch filter, homomorphic filter, spectral analysis,
   autocorrelation, cross-correlation, phase symmetry detector
8. Advanced Measurements: fractal dimension (box-counting), lacunarity,
   orientation analysis (structure tensor), granulometry, radial
   distribution, 3D intensity profiles

crates/ri-io/src/lib.rs

@@ -3,6 +3,8 @@ pub mod dicom;
 pub mod fits;
 pub mod nifti;
 pub mod raw_import;
+pub mod ome_tiff;
+pub mod bio_formats_metadata;
 
 use std::path::Path;