Byte order is not checked when creating a TensorProto with tf.make_tensor_proto() from a numpy.ndarray with non-native byte order. This leads to wrong tensor data.

I would expect that byte order is checked and swapped if necessary before assigning to tensor_proto.tensor_content in [python/framework/tensor_util.py](https://github.com/tensorflow/tensorflow/blob/v2.8.0/tensorflow/python/framework/tensor_util.py#L523)

x = np.ones(shape=(1, 2), dtype=np.float32)
x_bswap = x.astype('>f4') # assuming native byte order is little endian

y = tf.make_ndarray(tf.make_tensor_proto(x))
y_bswap = tf.make_ndarray(tf.make_tensor_proto(x_bswap))

print(y)
print(y_bswap)

assert np.array_equal(x, y)
assert np.array_equal(x, y_bswap)

[[1. 1.]]
[[4.6006e-41 4.6006e-41]]
---------------------------------------------------------------------------
AssertionError                            Traceback (most recent call last)
in <cell line: 7>()
      5 print(y_bswap)
      6 assert np.array_equal(x, y)
----> 7 assert np.array_equal(x, y_bswap)

AssertionError: