dedicated storedFields executor experiment

[magibney]

This is a very simple change (best viewed ignoring whitespace) that simply wraps the parts of the code that instantiate ThreadLocal storedFields and delegates them to a dedicated executor.

Consolidating in this way, and with an executor that allows its threads to die (at least for initial POC) should all but eliminate the "ThreadLocal storedFields cartesian product accumulation over time" problem, to hold us over until we have non-ThreadLocal storedFields in solr 9.7.

I determined these locations by analyzing GCP profiler (to identify code that in practice instantiates ThreadLocal storedFields).

[patsonluk]

Just curious! For the threadlocal usage, is it something buried deep in the lucene end (was reading this PR)? And that it's triggered by SolrIndexSearcher#doc?

Therefore your proposed change just invokes the doc within threads with shorter lifespan (now managed by storedFieldsExecute), so the threadlocals will not accumulate as badly?