fix: coerce numpy scalars in colmap write_next_bytes for NumPy 2.x

NumPy 2.x no longer implicitly converts a size-1 ndarray or a NumPy
scalar to a Python scalar inside struct.pack, so writing COLMAP binary
models (e.g. Point3D.error passed as np.array([0])) raised
'struct.error: required argument is not a float'. Coerce ndarray/generic
to a Python scalar via .item() before packing.

Fixes the failing tests/process_data/test_process_images.py cases.

nerfstudio/data/utils/colmap_parsing_utils.py

@@ -96,6 +96,11 @@ def write_next_bytes(fid, data, format_char_sequence, endian_character="<"):
     if isinstance(data, (list, tuple)):
         bytes = struct.pack(endian_character + format_char_sequence, *data)
     else:
+        # NumPy 2.x no longer implicitly converts a size-1 ndarray (or a NumPy
+        # scalar) to a Python scalar inside struct.pack, which raises
+        # "required argument is not a float/integer". Coerce explicitly.
+        if isinstance(data, (np.ndarray, np.generic)):
+            data = data.item()
         bytes = struct.pack(endian_character + format_char_sequence, data)
     fid.write(bytes)