Bouncner

Done for `benchmark_all.sh` with #2332.

> Or is it actually expected behavior? I would definitely say no. I am not even sure if we have ever considered using `runs` with multiple clients.

Yes, but not counting it as runs *per client* doesn’t make sense imo.

What about not supporting runs when the shuffled mode is used? Simply an assert and a message such as `The shuffled mode does not support limiting the number of benchmark...

And while we're at it: we could change the following output for shuffled runs `- Max duration per item is 2400 seconds`.

Reopening since one of the discussed issues (LZ4 performance) that led to this PR do not appear to be fixed. This is the current master on my MacBook. ``` BM_TableScanSorted/LZ4/0.001000/Int/Sorted...

In my great and unmatched wisdom, I consider this PR to be perfectly fine. The mentioned LZ4 issue is caused by a suboptimal implementation of iterator advance. In a future...

Short summary of the current status: Our current segment iterators remain input iterators. The reason is that we cannot return a real reference which is required from all iterator traits...

Absolutely loving it! Where can I approve?

I might get that wrong, but from my understanding this library saves us to manually add/ensure the required concepts for C++20 functions like std::lower_bound (or std::ranges::lower_bound) to be non-linear. BUT...