Release entryBufferPool once

[owen-d]

While investigating an issue, I found a bug where we can put the *entryBuffer back into the pool twice. This PR ensures we only release it once.

This PR also improves efficiency a bit by ensuring we don't nil the underlying slice unnecessarily, instead letting the pool reuse it.

This can happen when:
First, we put it into the pool here:

func (i *reverseEntryIterator) Next() bool {
	i.load()
	if len(i.buf.entries) == 0 {
		entryBufferPool.Put(i.buf)
		i.buf.entries = nil
		return false
	}
	i.cur, i.buf.entries = i.buf.entries[len(i.buf.entries)-1], i.buf.entries[:len(i.buf.entries)-1]
	return true
}

then it gets Get()'d by another goroutine, has it's entries appended, then the original goroutine puts it into the pool a second time here:

func (i *reverseEntryIterator) Close() error {
	if i.buf.entries != nil {
		i.buf.entries = i.buf.entries[:0]
		entryBufferPool.Put(i.buf)
		i.buf.entries = nil
	}
	if !i.loaded {
		return i.iter.Close()
	}
	return nil
}

[cyriltovena]

The comment is a bit weird. Should we just reslice when we get ?

[owen-d]

We could do this after Get'ing, yeah -- I believe that has equivalent results. The important part is that we don't nil it here because the reason we're using a pool is to preserve the underlying slice so we don't have to reallocate it. Does that make sense?

[cyriltovena]

Yeah I would just re-slice entries are never nil.

[cyriltovena]

LGTM

[cyriltovena]

Good catch btw.

We definitively lacking a test on close, after iteration finished.