perf: longer-range AC2_Hamiltonian improvements

[lkdvos]

This PR addresses some of the performance issues that got brought up recently for MPOs beyond nearest-neighbour.

1. BlockTensorMap slicing was repeated multiple times, which could just be done once upfront (minor overhead)
2. the "CA" term had a contraction order mistake, leading to an intermediate object of size D^2 V d^2 before the final D^2 d^4 instead of the (better) intermediate d^4 V.
3. the "AA" term was contracting terms that did not contribute anything.

A concrete example of this last machinery is found by considering a longer-range term that ends on the right physicalspace. This has support in the first A block, but then connects to the the second B block (and is already accounted for in the "AB" term). Nevertheless, this was still contributing a channel to the "AA" terms, needlessly increasing the MPO bond dimension on the most expensive contraction.

I already had some machinery in place for detecting something similar, but was missing quite a few cases still, which mostly bites for hamiltonians that have a variety of terms that span ranges >2.

[codecov]

Codecov Report

❌ Patch coverage is 75.00000% with 10 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
.../algorithms/derivatives/hamiltonian_derivatives.jl	75.00%	10 Missing ⚠️

Files with missing lines	Coverage Δ
.../algorithms/derivatives/hamiltonian_derivatives.jl	89.51% <75.00%> (-3.91%)	⬇️

... and 2 files with indirect coverage changes

🚀 New features to boost your workflow:

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