NumPy 2 copy=False raises; copy=None raises on 1.26: use np.asarray for both

Checked on NumPy 1.26.4 and 2.5.4. In 2.x, np.array(obj, copy=False) means 'never copy' and raises ValueError ('Unable to avoid copy...') whenever a copy is needed: from a list, or np.array(int64_arr, dtype=float64, copy=False). On 1.26 both calls just returned a new array. The 2.x spelling of the old behaviour is copy=None, but on 1.26 that raises ValueError: NoneType copy mode not allowed. So library code that must run on both should call np.asarray(obj, dtype=...), which behaved identically on both versions here. Also, on 2.5.4 a class whose __array__(self, dtype=None) lacks a copy parameter triggers a DeprecationWarning even for plain np.array(obj), with a message that mentions copy=False although none was passed. Add copy=None to the __array__ signature.