Fix higher_order_moment in the core and einsum tenalg backends - #650

tl.tenalg.higher_order_moment fails with both tenalg backends:

import numpy as np, tensorly as tl
x = np.random.rand(6, 3)
tl.tenalg.higher_order_moment(x, 3)
# core:   TypeError: batched_outer() takes 1 positional argument but 2 were given
# einsum: ValueError: einstein sum subscripts string includes output subscript 'c' multiple times

• core: batched_outer takes a list of tensors since the batched-tensordot API change, but higher_order_moment still called it with two arguments. Now it passes [moment, tensor].
• einsum: the equation used a single operand and repeated the feature subscripts in the output ("bc->cc..."). It now uses order copies of the tensor, each with its own feature subscripts and a shared batch subscript, and divides by the number of samples, e.g. "ab,ac,ad->bcd" for a matrix and order=3.

There were no tests for this function, so I added tensorly/tenalg/tests/test_moments.py. It compares against np.einsum for a matrix (orders 1 and 3) and for samples that are tensors themselves (order 2). The test fails on main with both backends and passes with this change. I also ran pytest tensorly/tenalg with TENSORLY_TENALG_BACKEND=core and =einsum (284 passed each) and black --check.