Optimize: export speed, fix scaling bug, and improve workflow

- Parallelized folder export for massive speedup

- Implemented exclusion mask caching

- Fixed statistics discrepancy by scaling exclusion coordinates

- Hardcoded CSV format to semicolon separator and comma decimal

- Defaulted file/folder dialogs to images/ directory

- Added unit test for coordinate scaling

tests/test_image_processor.py

@@ -87,3 +87,37 @@ def test_set_overlay_color():
     # invalid hex does nothing
     proc.set_overlay_color("blue")
     assert proc.overlay_r == 0
+
+def test_coordinate_scaling():
+    proc = QtImageProcessor()
+    
+    # Create a 200x200 image where everything is red
+    red_img_small = Image.new("RGBA", (200, 200), (255, 0, 0, 255))
+    proc.orig_img = red_img_small # satisfy preview logic
+    proc.preview_img = red_img_small 
+    
+    # All red. Thresholds cover all red.
+    proc.hue_min = 0
+    proc.hue_max = 360
+    proc.sat_min = 10
+    proc.val_min = 10
+    
+    # Exclude the right half (100-200)
+    proc.set_exclusions([{"kind": "rect", "coords": (100, 0, 200, 200)}])
+    
+    # Verify small stats
+    s_small = proc.get_stats_headless(red_img_small)
+    # total=40000, keep=20000, excl=20000
+    assert s_small.total_all == 40000
+    assert s_small.total_keep == 20000
+    assert s_small.total_excl == 20000
+    
+    # Now check on a 1000x1000 image (5x scale)
+    red_img_large = Image.new("RGBA", (1000, 1000), (255, 0, 0, 255))
+    s_large = proc.get_stats_headless(red_img_large)
+    
+    # total=1,000,000. If scaling works, keep=500,000, excl=500,000.
+    # If scaling FAILED, the mask is still 100x200 (20,000 px) -> excl=20,000.
+    assert s_large.total_all == 1000000
+    assert s_large.total_keep == 500000
+    assert s_large.total_excl == 500000