fix: Handle inhomogeneous array shapes in to_arrays()

When arrays have different shapes, np.array(values, dtype=object) can
still fail due to numpy's broadcasting behavior. Instead, create an
empty object array and assign elements individually.

This matches v0.14.7 behavior where blobs were extracted from structured
arrays with object dtype columns.

Fixes #1380

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: dimitri-yatsenko

src/datajoint/expression.py

@@ -1019,7 +1019,10 @@ def to_arrays(self, *attrs, include_key=False, order_by=None, limit=None, offset
                     arr = np.array(values)
                 except ValueError:
                     # Variable-size data (e.g., arrays of different shapes)
-                    arr = np.array(values, dtype=object)
+                    # Must assign individually to avoid numpy broadcasting issues
+                    arr = np.empty(len(values), dtype=object)
+                    for i, v in enumerate(values):
+                        arr[i] = v
                 result_arrays.append(arr)
 
             if include_key: