#4977 added np.linalg.multi_dot, before np.matmul existed. However, this only works for single 2D matrices.

It ought to be relatively straightforward to add a similar function that works on stacks of matrices, using np.matmul.

What should such a function be called? Or can multi_dot be internally changed to actually call matmul? Because for the inputs it currently accepts, there would be no visible difference in behaviour