Implement is_objectable_dt64, astype_array

[flexatone]

DTYPE_OBJECTABLE_DT64_UNITS = frozenset((
        'D', 'h', 'm', 's', 'ms', 'us',
        ))

def is_objectable_dt64(array: TNDArrayAny) -> bool:
    '''This function assumes a dt64 array.
    '''
    unit = np.datetime_data(array.dtype)[0]
    if unit not in DTYPE_OBJECTABLE_DT64_UNITS: # year, month, nanosecond, etc.
        return False
    # for all dt64 units that can be converted to object, we need to determine if the can fit in the more narrow range of Python datetime types.
    years = array[~np.isnat(array)].astype(DT64_YEAR).astype(DTYPE_INT_DEFAULT) + 1970
    if np.any(years < datetime.MINYEAR):
        return False
    if np.any(years > datetime.MAXYEAR):
        return False
    return True

def is_objectable(array: TNDArrayAny) -> bool:
    '''If an array is dt64 array, evaluate if it can go to Python object without resolution loss or other distortions (coercion to integer).
    '''
    if array.dtype.kind in DTYPE_NAT_KINDS:
        return is_objectable_dt64(array)
    return True


def astype_array(array: TNDArrayAny, dtype: TDtypeAny | None) -> TNDArrayAny:
    '''This function handles NumPy types that cannot be converted to Python objects without loss of representation, namely some dt64 units. NOTE: this does not set the returned array to be immutable.
    '''
    dt = np.dtype(None) if dtype is None else dtype
    dt_equal = array.dtype == dt

    if dt == DTYPE_OBJECT and not dt_equal and array.dtype.kind in DTYPE_NAT_KINDS:
        if not is_objectable_dt64(array):
            # NOTE: this can be faster implemented in C
            post = np.empty(array.shape, dtype=dt)
            for iloc, v in np.ndenumerate(array):
                post[iloc] = v
            return post
    if dt_equal and array.flags.writeable is False:
        # if dtypes match and array is immutable can return same instance
        return array
    return array.astype(dt)