We currently have benchmarks for a keys-only batched top-k algorithm. Since, the most frequent workloads require outputting the index along with the key, we want to have benchmarks covering this use case.

The plan is to implement indexed top-k by passing in a counting_iterator per segment. We can later on still cover

This issue can be closed when we have a benchmark for indexed top-k that outputs the indices (within the segment) along with the keys, where we want to limit indices to i32 for now. Later potentially extending to i16 (e.g., for small segments) and i64.