fix(prune): avoid extra iterator consumption

[Forostovec]

What was wrong:

Double consumption on early exit: the for key in &mut keys loop consumes the next item before the body runs. If the limiter triggers at loop start, we break, but the current key is already consumed. The subsequent keys.next().is_none() after the loop consumed yet another key.
Misleading done flag: done was derived by advancing the iterator after the loop, not by whether the loop actually exhausted it.

Why this happens

for implicitly calls next() before each iteration.

The extra next() to determine done advanced the iterator once more, skipping work and misreporting completion.

What changed

1. Loop control:

• Replaced for key in &mut keys with an explicit loop that checks limiter.is_limit_reached() before calling keys.next().

2. Compute done within the loop:

• done = true only when keys.next() returns None.
• On limit reach, break without advancing and set done = false.

[mattsse]

this looks correct

I have one style suggestion

[mattsse]

can we changes this so that we use .peekable on keys.into_iter().peekable()

and then

while keys.peek().is_some() {
     if limiter.is_limit_reached() {
        ...
    key = keys.next()
}

[shekhirin]

agreed, because I think with these changes we also incorrectly set done = false; if both the limit is reached and there's no more keys left to process.

[Forostovec]

can we changes this so that we use .peekable on keys.into_iter().peekable()

and then

while keys.peek().is_some() {
     if limiter.is_limit_reached() {
        ...
    key = keys.next()
}

Like this? corrected

[shekhirin]

@mattsse maybe we can even use https://doc.rust-lang.org/std/iter/struct.Peekable.html#method.next_if here?

[mattsse]

yeah that's also an option @Forostovec

[codspeed-hq]

CodSpeed Performance Report

Merging #19758 will not alter performance

_{Comparing Forostovec:fix/prune-iterator-done-overconsumption (b43a25a) with main (cac2443)}

Summary

✅ 81 untouched

[mattsse]

this lgtm, ty!