Correlation functions for MPSEnvironment

[jhauschild]

The MPS has a few functions for evaluating correlation functions of the form C(j) = <A(i) B(i+j)> == <psi| A(i) B(i+j) |psi>. All of these things can also be evaluated between different MPS, C2(j) = <phi| A(i) B(i+j) |psi>.
In terms of code, this means to mostly copy the methods from the MPS, properly insert get_LP and get_RP of the environments, and use the B.conj() and one set of S from the phi instead of the psi - the tensor network contraction itself is completely analogous.

One further, slightly tricky thing is that we need to obtain dominant left/right eigenvectors of the MPSTransferMatrix to get initial LP/RP. It's probably best to do this in a class method of the MPSEnvironment, e.g. from_TM_eigenvectors.

Requested in https://tenpy.johannes-hauschild.de/viewtopic.php?t=276

[jhauschild]

This was implemented in #267, up to the from_TM_eigenvectors which is still missing.

[jhauschild]

Realizing now that the eigenvectors of the MPSEnvironment(psi, psi) are trivial identities, so this already works.
On the other hand, MPSEnvironment(psi, phi) for phi != psi has dominant TM eigenvalues < 1, so there's no point in calculating this (for infinite systems - for finite, on the hand, just MPSEnvironment(psi, phi) initialized the correct thing.
So overall we have everything we need (that makes sense).