Michael McCandless

This looks really awesome @markrmiller -- we are perpetually in need of better benchmarking tools!

Is this change still relevant? Or did we achieve laziness on subset of stored fields in a different way maybe? Thanks @JoeHF! > no obvious regression or perf improvement, guess...

Thank you for persisting so hard on this one @shahrs87 -- I'm sorry it looks like we should close it at this point, but your efforts / iterations were needed...

Somewhat related to [this newish issue](https://github.com/apache/lucene/issues/12740) (how to configure concurrent HNSW graph building). Let's stick with the straightforward "pass args to ctor" for configuring Codec component writers?

It looks like this PR is a nice improvement to MLQ quality, and we agree we should just enable it by default (`Similarity` can turn it off if the old...

@dweiss is this still relevant? It looks like the original issue was hard to repro too...

It looks like we are abandoning this idea -- too much new API surface area added?

> > Yeah, those are actually API bugs? > > They do look like API issues to me. Useful warning, by the way. It's awesome that this change uncovered such...

Thanks @wjp719. It looks like this is a nice opto for narrow use cases and the concern is adding a new API, especially to such a hot class as `PostingsEnum`,...

Hi @zacharymorn -- this change is awesome! The world of servers has rapidly become massively concurrent and Lucene has (generally) been slow to adopt it. I like this hardish switch...