Preprocessing: Implement Q-matrix permutation

[iagoleal]

The linear nature of MPS makes them most appropriate for tridiagonal matrices (1d neighbor connections). We can permute the site indices to better approximate a 1d system.

This is itself an NP problem, but perhaps the ideas here could help (Appendix III.A.4):
https://arxiv.org/pdf/2403.00910

[bernalde]

I can see why the permutation is NP, as solving tridiagonal QUBOs with TensorNetworks is in P https://arxiv.org/abs/2309.10509
In the reference you provide there are some references, but their algorithm is the most straightforward in my opinion