Reverse entry iterator pool

[cyriltovena]

The iter.NewReversedIter was causing a lot of allocs. This uses a pool and remove the labels string since we never needed it.

benchcmp:

benchmark                        old ns/op     new ns/op     delta
BenchmarkBackwardIterator-16     15943073      14537418      -8.82%

benchmark                        old allocs     new allocs     delta
BenchmarkBackwardIterator-16     47217          47225          +0.02%

benchmark                        old bytes     new bytes     delta
BenchmarkBackwardIterator-16     27444423      23664350      -13.77%

[owen-d]

Overall, lgtm once you get your tests passing. I'm a little wary of having implementations which only satisfy the EntryIterator ifc in compilation, but not in reality (returning "" for Labels()). That being said, I see this isn't the only place we do this, so I don't want to block it.

I think the long path forward may be using subset/superset interfaces i.e. type LabeledEntryIterator interface { Labels() string, EntryIterator }, but that type of refactor is way too much for this PR.

What about an adapter like

type BoundLabelsIterator struct{
  ls string
  EntryIterator
}

func (i BoundLabelsIterator) Labels() string {return i.ls}

?

[cyriltovena]

Overall, lgtm once you get your tests passing. I'm a little wary of having implementations which only satisfy the EntryIterator ifc in compilation, but not in reality (returning "" for Labels()). That being said, I see this isn't the only place we do this, so I don't want to block it.

I wasn't sure how to do this properly, I just catched that with labels it's a bit slower, not much but with the amount of those struct allocated it can make a difference.