Change remove from swap to shift in index map

[mustafasrepo]

Which issue does this PR close?

Closes #.

Rationale for this change

In the #PR9034, @viirya resolved CI error, by replacing .remove method of IndexMap with .swap_remove (they are equal).
However, while reviewing I thought maybe it is better to use .shift_remove(where relative order is preserved after removal).

As an example, assume we have the alternative valid orderings: [d ASC, b ASC], [e ASC, f ASC], [d ASC, h ASC],
and find_longest_permutation is called with expressions: d, h, e, f, b. Conceptually any result in the following form:
[permutation(d ASC, e DESC) + permutation(b ASC, f ASC, h DESC)] is valid. Meaning,
following orderings:
[d ASC, e ASC, b ASC, f ASC, h ASC],
[e ASC, d ASC, b ASC, f ASC, h ASC],
[d ASC, e ASC, f ASC, b ASC, h ASC],
...
are all valid.

However, to not depend on any internal implementation. It is better to prefer argument ordering between permutations. e.g we want to return ordering: [d ASC, e ASC, b ASC, f ASC, h ASC], for the arguments d, h, e, f, b.

Similarly, we want to return ordering: [e ASC, d ASC, b ASC, f ASC, h ASC], for the arguments e, d, h, f, b to be consistent, and to be least surprising.
Replacing .swap_remove with .shift_remove ensures this property.

What changes are included in this PR?

Are these changes tested?

Yes

Are there any user-facing changes?

[alamb]

Thank you @mustafasrepo -- this makes sense to me

I also double checked the indexmap docs and maintaing the order of the removed item makes sense to me
https://docs.rs/indexmap/latest/indexmap/map/struct.IndexMap.html#method.swap_remove