Thread pool size should not be a global (cluster update API) setting

[mikemccand]

Spinoff from #15585

Today you can e.g. set threadpool.bulk.size across the cluster, but this is dangerous, because it really should be node level setting instead (maybe you have "hot" nodes with many cores, and "cold" nodes with fewer cores).

[jasontedor]

Closed in favor of #18613