fix(__array__): coerce to numpy + accept copy kwarg (numpy 2.x strict mode)

NumPy >=1.26 raises "object __array__ method not producing an array"
when __array__ returns a non-np.ndarray. Our impl returned self._array
directly, but _array can be a jax.Array on the JAX path (e.g. set by
`_array = jnp.where(...)` in __setitem__). Coerce to numpy via
np.asarray to satisfy the protocol contract.

Also added the `copy` kwarg introduced in numpy 2.0; without it, future
numpy may pass copy positionally and break the call.

Surfaced by python_matrix.yml smoke runs on autogalaxy_workspace, which
hit `np.array(vis.slim)` in autoarray/plot/utils.py:plot_visibilities_1d
on every Python version. Issue #74 in PyAutoBuild Category A.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

Author: 2 people

autoarray/abstract_ndarray.py

@@ -418,10 +418,20 @@ def __repr__(self):
             self.__class__.__name__,
         )
 
-    def __array__(self, dtype=None):
-        if dtype:
-            return self._array.astype(dtype)
-        return self._array
+    def __array__(self, dtype=None, copy=None):
+        # numpy >=1.26 requires __array__ to return an actual np.ndarray
+        # ("object __array__ method not producing an array" otherwise),
+        # and numpy 2.0 added the `copy` kwarg. self._array can be a
+        # jax.Array on the JAX path (set in `_array = jnp.where(...)` etc.)
+        # — coerce to numpy so the protocol contract holds.
+        arr = self._array
+        if not isinstance(arr, np.ndarray):
+            arr = np.asarray(arr)
+        if dtype is not None:
+            arr = arr.astype(dtype)
+        if copy:
+            arr = arr.copy()
+        return arr
 
     def __len__(self):
         return len(self._array)