Make an eigendecomp function

[mtfishman]

To complement the new eigen interface introduced in #334, it may be nice to have another function, possibly call eigendecomp, that returns a factorization of an ITensor in terms of left and right eigenvectors and the eigenvalues:

i = Index(2)
A = randomITensor(i', i)
V, D, W = eigendecomp(A)
A == V * D * W

Here, V stores the right eigenvectors of A and W stores the left eigenvectors. In the case of a Hermitian A, W would be related to the complex conjugation of V. More generally, W is related to the matrix inverse of V.

This could mostly be written as a wrapper around eigen, where in the non-Hermitian case it could do two calls to eigen to get the left and right eigenvectors (and do some extra normalization of the sets of left and right eigenvectors).

I think it is useful to have both eigen and eigendecomp, where sometimes just the right or left eigenvectors are desired, while other times it is nice to have a full factorization that can easily reproduce the original ITensor (for example think about using it in a tensor renormalization algorithm like TRG or CTMRG, where you want to get a factorized replacement for a tensor in the network).

[emstoudenmire]

Sounds like a great idea