Revise fix for integration tests

[Krzmbrzl]

This (effectively) reverts 445eeae as we
do indeed want the semantics to be such that symmetrizers need to be
constructed with swapped bra/ket indices compared to what
external_indices(…) returns. The underlying issue in the integration
tests was that external_indices was called on the symmetrizer alone
rather than the entire expression, in which case it currently has the
opposite behavior in terms of bra/ket ordering.

Hence, the correct way to fix the integration tests is to use
external_indices in the way it was intended to be used and leave the
construction of the symmetrizer tensors as it was.

See also #471

[evaleev]

@Krzmbrzl this is fine, but points out the flaw with what external_indices returns currently. Bra/ket/aux are properties of slots, not indices. Now, you may think that each external index should only appear in 1 slot, and indeed only in such situations external_indices makes sense. What we want this function to return, however, would be {index,slot} pairs, the latter would allow us to support aux slots, and would make get_{bra,ket}_idx safer/cleaner.

[Krzmbrzl]

What we want this function to return, however, would be {index,slot} pairs, the latter would allow us to support aux slots, and would make get_{bra,ket}_idx safer/cleaner.

So you mean something like { "i_1", Slot::Bra}?

Sounds like an excellent idea 👍