Add NTU bond environments

[Yue-Zhengyuan]

This PR adds the construction of bond environments using neighboring tensors and Schmidt weights of a bond. They are intended to be used in Neighborhood Tensor Update for InfiniteWeightPEPS. The code is adapted from YASTN, and I have put the credit in src/algorithms/contractions/bondenv/benv_ntu.jl.

TODO

• The direction names t/l/b/r are directly borrowed from YASTN, and need to be changed to north, west, south, east to follow PEPSKit convention after everything else have been figured out.
• Determine the best codomain/domain choice for the PEPSOrths (the X and Y tensors) that can reduce the number of permutes and twists.
• enlarge_corner_se, enlarge_corner_nw may reuse functions in existing CTMRG code.

[codecov]

Codecov Report

Attention: Patch coverage is 88.77551% with 11 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/algorithms/contractions/bondenv/benv_tools.jl	81.66%	11 Missing ⚠️

Files with missing lines	Coverage Δ
src/PEPSKit.jl	100.00% <ø> (ø)
src/algorithms/contractions/bondenv/benv_ntu.jl	100.00% <100.00%> (ø)
src/algorithms/contractions/bondenv/benv_tools.jl	81.66% <81.66%> (ø)

🚀 New features to boost your workflow:

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

[Yue-Zhengyuan]

The NTU bondenvs can also be constructed for InfinitePEPS without Schmidt weights on bonds (this is the approach in YASTN). But I found that NTU works better (at least for Heisenberg model) if I include the weights. At $D = 4$ with NTUEnvNN, the NTU energy for InfinitePEPS is -0.66858; for InfiniteWeightPEPS it can be further improved a bit to -0.66885.

I think the reason is quite simple. For InfinitePEPS, the NTUEnvNN bondenv is (I only draw the ket layer without physical axes; the open ends are contracted with the bra layer)

            |   |
        ----T---T----
            |   |
    ---T---()   ()---T---
            |   |
        ----T---T----
            |   |

For InfiniteWeightPEPS with bond weights, the NTUEnvNN bondenv is (assume the weights have been absorbed into the tensors)

                |       |
                √λ      √λ
                |       |
        ---√λ---T-------T---√λ---
                |       |
    ---√λ---T---()      ()---T---√λ---
                |       |
        ---√λ---T-------T---√λ---
                |       |
                √λ      √λ
                |       |

So the bondenv has an additional circle of $\sqrt{\lambda}$'s, which is intuitively a more accurate than the bondenv for InfinitePEPS. This is why I modified the NTU in YASTN to work with InfiniteWeightPEPS instead of InfinitePEPS.

[Yue-Zhengyuan]

Edit: I'll soon add an option to control whether to further absorb the additional circle of weights when constructing the NTU bond environment.