Refactor depth reference handling and cleanup

- Unified reference depth calculation to use a single implementation in common/depth.py.
- Removed server-side reference depth computation and adjusted depth.json to only include image_size and intrinsics.
- Updated depth texture handling to ignore transparent, invalid, and non-positive pixels when calculating reference depth.
- Changed reference depth calculation to use the 95th percentile of valid depth values.
- Adjusted depth direction fitting to return a flat direction when no samples are available.
- Removed cleanup threshold parameters from Cutout and related nodes, simplifying the interface.
- Updated tests to verify new reference depth logic and ensure compatibility with existing functionality.

src/anyimage/operators/cutout_tool/geometry.py

@@ -464,7 +464,6 @@ def boundary_vertices_from_faces(vertex_count, faces):
 
 
 CUTOUT_ALPHA_THRESHOLD = 0.9
-CUTOUT_CLEANUP_UI_SCALE = 100.0
 
 
 @dataclass(frozen=True)
@@ -532,17 +531,6 @@ def world_plane_metric(matrix_world):
     return ((x.dot(x), x.dot(y)), (x.dot(y), y.dot(y))), z.length
 
 
-def cutout_cleanup_value(spacing, world_metric):
-    import numpy as np
-
-    if spacing <= 0.0:
-        raise ValueError("Cutout point spacing must be greater than zero")
-    area_scale = abs(float(np.linalg.det(metric_transform(world_metric))))
-    if area_scale <= 1e-12:
-        raise ValueError("The Image Empty has a degenerate world transform")
-    return CUTOUT_CLEANUP_UI_SCALE * float(spacing) ** 2 / area_scale
-
-
 def counter_clockwise_faces(points, faces):
     import numpy as np
 
@@ -620,27 +608,13 @@ def effective_cutout_edge_length(
     )
 
 
-def selection_calibration(
-    content_values, target_shape, threshold=CUTOUT_ALPHA_THRESHOLD
-):
-    """Return Cutout validity and transparency ratio at geometry resolution."""
+def build_reference_mask(content_values, image_size):
+    """Resample Cutout content into a boolean mask at Depth texture resolution."""
     import numpy as np
 
     values = np.asarray(content_values, dtype=np.float32)
-    target_height, target_width = (int(value) for value in target_shape)
-    source_height, source_width = values.shape
-    y = np.minimum(
-        (np.arange(target_height) * source_height / target_height).astype(np.int64),
-        source_height - 1,
-    )
-    x = np.minimum(
-        (np.arange(target_width) * source_width / target_width).astype(np.int64),
-        source_width - 1,
-    )
-    resized = values[y[:, None], x[None, :]]
-    valid = (resized > 0.0) & (resized >= float(threshold))
-    transparent_fraction = float(np.count_nonzero(~valid) / valid.size)
-    return valid, transparent_fraction
+    target_width, target_height = (int(value) for value in image_size)
+    return _resize_values(values, (target_width, target_height)) > 0.5
 
 
 def build_base_shape(