Update correlation length normalization

[lkdvos]

This should resolve #199 by correctly normalizing the transfer spectra.

@ogauthe, do you happen to have some explicit results that I can use to add tests?

[ogauthe]

The correlation length for the AKLT spin 2 wavefunction is slightly above 2. It may be added to https://github.com/QuantumKitHub/PEPSKit.jl/blob/master/test/ctmrg/correlation_length.jl
Getting a precise value is not very easy as the corner singular values decay super fast and the measured value is affected by the cutoff.
https://journals.aps.org/prb/abstract/10.1103/PhysRevB.96.121115 finds xi_AKLT = 2.065 using the boundary state of a semi infinite cylinder with.
Using an asymmetric CTMRG with a cutoff of 1e-13, I converge at chi=48 with xi_AKLT = 1.976
Using an asymmetric CTMRG with a cutoff of 1e-20, I converge at chi~430 with xi_AKLT = 2.004 (quite sensitive to details)
With a C4v symmetric CTMRG with chi=104, I find 2.010
With a C4v symmetric CTMRG with chi=202, I find 2.028

I guess "between 1.9 and 2.1" should do the trick

[codecov]

Codecov Report

Attention: Patch coverage is 70.00000% with 9 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/environments/ctmrg_environments.jl	37.50%	5 Missing ⚠️
src/states/infinitepeps.jl	0.00%	4 Missing ⚠️

Files with missing lines	Coverage Δ
src/algorithms/toolbox.jl	98.26% <100.00%> (+0.18%)	⬆️
src/states/infinitepeps.jl	71.13% <0.00%> (-3.06%)	⬇️
src/environments/ctmrg_environments.jl	67.91% <37.50%> (-1.36%)	⬇️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[ogauthe]

As discussed, I find it slightly misleading that by default the returned correlation may not be the largest one. Especially that it depends on which symmetry is implemented: in the AKLT example, correlation_length(ψ, env) returns 2 different results depending whether ψ makes use of U(1) or SU(2). As a user, I would expect that it returns the largest one by default, then sector is an advanced mode with finer control. Of course I am not using MPSKit so I may be unaware of established conventions.

[lkdvos]

@ogauthe, Can you open an issue for this? Something like correlation length interface discussion seems look a good idea, since I'm also supremely unhappy with the current output of PEPSKit's correlation length, since it is completely breaking from what MPSKit is returning (a single number).

We can merge this as "a fix" without altering the conventions, and then re-assess based on the collective discussion, since I would then also want the same behavior in MPSKit.

[pbrehmer]

Looks good to me! And I completely agree that MPSKit and PEPSKit should share the same correlation length interface. But more on that in #216...