LQ decomposition

[manuschneider]

If I use linalg.Qr, then the index direction is always from the Q tensor to the R tensor. I want to have it the other way round for an algorithm. This is even more important when Q or R would have only incoming or outgoing bonds, which is technically not allowed in Cytnx (but still works in the current implementation, see #497).

I found one solution that does not change the direction of a single bond (the latter is not compatible with symmetries):

1. Transpose the initial tensor (which redirects ALL bonds)
2. Do the Qr
3. Transpose Q and R

This is, however, not so clean to me. Ideally, we do not want to change the directions of indices if there is not a good reason for it (with Dagger = conjugate transpose being the only good reason I come up with).

There would be 2 possibilities to fix this:
i) Implement Lq: M = L * Q with bond direction from Q to L. Then, one could permute the indices of the initial tensor before using Lq and have the effect needed, a QR decomposition with reversed internal index. This idea works for Svd already: one can permute the initial indices and exchange the meaning of U and Vt.
ii) Have an additional argument in tensor decomposition functions (Qr, Svd, Eig, ...) which defines if the direction is from left to right or the other way round