ast/term: sort set's keys slice on insert

[srenatus]

Before, we'd sort a set's keys slice in-place, and only when it
was actually compared to another set. This side-effect of a comparison
can be unexpected.

Now, we'll keep the keys slice sorted by inserting every new element
into its sorted position.

Note that we still sort in-place on Compare, because its possible the
sets values have been altered in different ways (e.g. test TestSetCopy).

Analogous to #3823, where we did this with objects' keys slices.

---

Also noticed a problem of #3823: the ordering could me messed up in ways different from inserting key/val pairs, e.g. by changing key names in Iter(), so we'll need to care about our invariant after the Iter() call.

---

🚧 Some discussion items below 👇

[srenatus]

We cannot get rid of the other sort calls,

• (set).Sorted(): https://github.com/open-policy-agent/opa/blob/v0.34.2/ast/term.go#L1524
• (set).Compare(...): https://github.com/open-policy-agent/opa/blob/v0.34.2/ast/term.go#L1371-L1372

because we're returning the underlying keys slice in (set).Slice() []*Term. If we made that return a copy instead, we would be able to ensure that our ordering isn't messed up.

[srenatus]

I've looked more into this, but returning a copy of the terms from Slice() is at odds with how the other methods work: none of the Object methods, for example Keys(), or Iter(), operate on copies (except Map). So, this would single out sets in the wrong way.

[srenatus]

OK, I've dialled this back a little: we now do what's advertised in the title: sort sets at insertion time. We still sort them (again) when comparing, or when printing them, because we don't know the ordering hasn't been messed with.

[tsandall]

I think we should just make this PR do the sort-on-insert change to sets. The reason is that the data structures do not support in-place mutation today--if something reaches into the elements and mutates them, we're going to get undefined behaviour with Iter() and other functions. Unfortunately, sort-on-compare does not solve this. Let's yank those other changes out and just sort-on-insert in sets.

[tsandall]

LGTM

[srenatus]

@tsandall sorry f-pushed another one after your review -- this was the change I had not included before 👇