perf: avoid unnecessary large allocations

[ariel-miculas]

Which issue does this PR close?

Related a bit to #22526

Needs rebasing once #22416 is merged

Rationale for this change

split_off does this:

Returns a newly allocated vector containing the elements in the range [at, len). After the call, the original vector will be left containing the elements [0, at) with its previous capacity unchanged.

which is bad when taking a small slice from a large Vec, for two reasons:

• it will allocate memory for the remaining elements, which are a lot more than n
• it will return a Vec with a very large capacity compared to its length

split_vec_min_alloc still has some issues: #22548 but it uses drain + collect when n is small, which is better because it only allocates for the initial n elements and doesn't inflate the capacity

What changes are included in this PR?

Are these changes tested?

Yes

Are there any user-facing changes?

No

[kosiew]

@ariel-miculas
Thanks for working on this.
I do not see any blocking issues, just one small cleanup suggestion.

[kosiew]

Small cleanup idea: since this helper is only used by one test assertion, I think we can avoid adding the test-only values_capacity API.

The child tests module should be able to read the private field directly, so this could be simplified to:

let capacity_before = gv.values.capacity();// ...assert_eq!(gv.values.capacity(), capacity_before, ...);

That would let us remove the #[cfg(test)] impl without changing behavior.

[ariel-miculas]

thanks, applied your suggestion