fix: always use min-channel and add grayscale-only OCR path (refs #113)

Two fixes:
1. Always use min-channel for color images instead of gated comparison
   that was falling back to standard grayscale (which has only 23%
   contrast for white-on-green VIN stickers).
2. Add grayscale-only OCR path (CLAHE + denoise, no thresholding)
   between adaptive and Otsu attempts. Tesseract's LSTM engine is
   designed to handle grayscale input directly and often outperforms
   binarized input where thresholding creates artifacts.

Pipeline order: adaptive threshold → grayscale-only → Otsu threshold

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

ocr/app/preprocessors/vin_preprocessor.py

@@ -167,20 +167,20 @@ class VinPreprocessor:
         b_channel, g_channel, r_channel = cv2.split(bgr_image)
 
         min_channel = np.minimum(np.minimum(b_channel, g_channel), r_channel)
+        gray = cv2.cvtColor(bgr_image, cv2.COLOR_BGR2GRAY)
 
         min_std = float(np.std(min_channel))
-        gray = cv2.cvtColor(bgr_image, cv2.COLOR_BGR2GRAY)
         gray_std = float(np.std(gray))
+        logger.debug(
+            "Channel contrast: min-channel std=%.1f, grayscale std=%.1f",
+            min_std, gray_std,
+        )
 
-        # Use min-channel when it provides meaningfully more contrast
-        if min_std > gray_std * 1.1:
-            logger.debug(
-                "Using min-channel (std=%.1f) over grayscale (std=%.1f)",
-                min_std, gray_std,
-            )
-            return min_channel
-
-        return gray
+        # Always use min-channel for VIN images.  White text keeps
+        # min(B,G,R)=255 while any colored background drops to its
+        # weakest channel.  For neutral images the result is equivalent
+        # to grayscale, so there is no downside.
+        return min_channel
 
     def _apply_clahe(self, image: np.ndarray) -> np.ndarray:
         """