feat: Implement ONNX model inference for BEN2, Depth Anything 3, and MoGe

- Added `ben2_onnx.py` for BEN2 ONNX inference with preprocessing.
- Introduced `da3_onnx.py` for Depth Anything 3 inference, including camera projection calculations.
- Created `moge_onnx.py` for MoGe-2 ONNX inference and post-processing.
- Implemented shared ONNX Runtime session helpers in `onnx_runtime.py`.
- Added model synchronization script `sync_models.py` for downloading and verifying ONNX models.
- Developed unit tests for BEN2, Depth Anything 3, and MoGe ONNX models to ensure functionality.
- Updated `vendor.lock.toml` to remove unused artifacts.

tests/test_ben2_onnx.py

@@ -0,0 +1,53 @@
+import sys
+import unittest
+from pathlib import Path
+from types import SimpleNamespace
+
+import numpy as np
+from PIL import Image
+
+
+SOURCE_DIR = Path(__file__).resolve().parents[1] / "src" / "snapform"
+sys.path.insert(0, str(SOURCE_DIR))
+
+import ben2_onnx
+
+
+class FakeSession:
+    def get_inputs(self):
+        return [SimpleNamespace(name="input")]
+
+    def get_outputs(self):
+        return [SimpleNamespace(name="alpha")]
+
+    def run(self, output_names, feeds):
+        self.output_names = output_names
+        self.feeds = feeds
+        gradient = np.linspace(0.0, 1.0, 1024, dtype=np.float32)
+        return [np.tile(gradient, (1024, 1))[None, None]]
+
+
+class Ben2OnnxTest(unittest.TestCase):
+    def test_preprocess_matches_model_layout(self):
+        image = Image.new("RGB", (40, 20), (64, 128, 192))
+        result = ben2_onnx.preprocess(image)
+
+        self.assertEqual(result.shape, (1, 3, 1024, 1024))
+        self.assertEqual(result.dtype, np.float32)
+        self.assertTrue(result.flags.c_contiguous)
+
+    def test_infer_restores_original_size_and_alpha(self):
+        session = FakeSession()
+        image = Image.new("RGB", (40, 20), (10, 20, 30))
+
+        result = ben2_onnx.infer(session, image)
+
+        self.assertEqual(result.mode, "RGBA")
+        self.assertEqual(result.size, image.size)
+        self.assertEqual(session.feeds["input"].shape, (1, 3, 1024, 1024))
+        alpha = np.asarray(result.getchannel("A"))
+        self.assertLess(int(alpha[:, 0].max()), int(alpha[:, -1].min()))
+
+
+if __name__ == "__main__":
+    unittest.main()