Deprecated `pinv2` function in `scipy.linalg` about procrustes

[PaulWAyers]

I think there is no choice here. It bothers me that under-the-hood the SVD is the iterative SVD that can fail due to "lack of convergence" and this is a problem we've observed happens sometimes, as opposed to the rigorous factorization of a matrix which is always fine. But with current versions, scipy.pinv, scipy.pinv2, and numpy.pinv are all the same. No need to distinguish.

See the discussion scipy/scipy#8861 (comment)

[PaulWAyers]

We need to remember this: if we end up with "SVD did not converge" errors we will need to add in a brute strength implementation using the dgesvd SVD driver (accessible in scipy).