remove final permutedims() in binary_einsum

[starsfordummies]

maybe I don't fully understand the logic in all those index sets in binary_einsum, but if we can get rid of the final permutedims() that would give a significant speedup. The tests hopefully will help see if I'm wrong

[jofrevalles]

Oh, so we could put something that reorders the indices instead of the inner array, right?

[mofeing]

I'm thinking that we can probably better avoid kwargs and directly use the ScopedValue as a performance configuration. Tell me if it's difficult for you and I can do it.

[mofeing]

Suggested change

[mofeing]

Oh, so we could put something that reorders the indices instead of the inner array, right?

yeah, kind of. we wouldn't really force an order doing permutedims on the array, but the other way around. ofc, this would be optional and would be enabled it by default.

[starsfordummies]

I still don't know if it makes sense to add a global option to force contract to return indices in a given order, when in most cases it decides the order by itself - know what I mean?
The way I implemented it right now it should force the order just when the out=.. list is given (which I understand is when you want to get a specific order of the outgoing legs, right? otherwise I don't understand what it's for)