Use pyvips for robust H&E thumbnail generation

Replace custom tifffile/PIL implementation with pyvips (libvips):
- Works with both OME-TIFF (convert=true) and original formats like SVS (convert=false)
- Handles RGB chunky format from raw2ometiff --rgb flag
- Production-ready library designed for whole slide imaging
- Efficiently handles pyramidal images automatically
- Eliminates complex array shape manipulation code
- More maintainable and reliable

Co-authored-by: adamjtaylor <14945787+adamjtaylor@users.noreply.github.com>

Author: Copilot

modules/make_miniature.nf

@@ -17,65 +17,17 @@ process make_miniature {
     """
     #!/usr/bin/env python
 
-    import tifffile
-    from PIL import Image
-    import numpy as np
+    import pyvips
 
-    # Open the OME-TIFF file and read the base (full resolution) image
-    # Let PIL handle the resizing via thumbnail() - it's more robust
-    with tifffile.TiffFile('$image') as tif:
-        # Always read from the base series (full resolution)
-        # Don't try to use pyramid levels as they may be too small or have odd shapes
-        if len(tif.series) > 0:
-            img_array = tif.series[0].asarray()
-        else:
-            img_array = tif.asarray()
+    # Use pyvips for robust thumbnail generation from any whole slide format
+    # Supports SVS, OME-TIFF, and other formats efficiently
+    image = pyvips.Image.new_from_file('$image', access='sequential')
     
-    # Handle array shape - squeeze out singleton dimensions
-    img_array = np.squeeze(img_array)
+    # Create thumbnail with max dimension of 512px
+    thumbnail = image.thumbnail_image(512)
     
-    # Ensure we have at least 2D array
-    if img_array.ndim < 2:
-        # After squeezing, if we end up with 1D, reshape to 2D
-        img_array = img_array.reshape(1, -1)
-    
-    # Handle different array shapes
-    # Expected shapes: (Y, X), (Y, X, 3), (Y, X, 4) for grayscale, RGB, or RGBA
-    if img_array.ndim > 3:
-        # If more than 3 dimensions after squeeze, take first slice
-        # This handles cases like (C, Y, X, S) -> take first channel
-        img_array = img_array[0]
-        img_array = np.squeeze(img_array)
-        # Check again after squeezing
-        if img_array.ndim < 2:
-            img_array = img_array.reshape(1, -1)
-    
-    # Ensure proper data type for PIL (uint8)
-    if img_array.dtype != np.uint8:
-        # Scale to uint8 range if needed
-        img_min = img_array.min()
-        img_max = img_array.max()
-        if img_max > 255:
-            # Avoid division by zero
-            if img_max > img_min:
-                img_array = ((img_array - img_min) / (img_max - img_min) * 255).astype(np.uint8)
-            else:
-                img_array = np.zeros_like(img_array, dtype=np.uint8)
-        else:
-            img_array = img_array.astype(np.uint8)
-    
-    # Convert numpy array to PIL Image
-    thumb = Image.fromarray(img_array)
-    
-    # Create thumbnail (maintains aspect ratio)
-    # PIL's thumbnail is efficient and handles large images well
-    thumb.thumbnail((512, 512))
-    
-    # Ensure RGB mode
-    if thumb.mode != "RGB": 
-      thumb = thumb.convert("RGB")
-    
-    thumb.save('miniature.jpg')
+    # Save as JPEG
+    thumbnail.write_to_file('miniature.jpg')
     """
   } else {
     """