Reduce the number of classes extending DocIdSetIterator

[jpountz]

Megamorphism of call sites to DocIdSetIterator#nextDoc and DocIdSetIterator#advance hurts search performance. See e.g. annotation HJ at https://benchmarks.mikemccandless.com/OrHighHigh.html where reducing polymorphism helped improve QPS by 40%. More specialization could also help, but before we start going this route, we should reduce the number of classes that extend DocIdSetIterator so that specialization is less needed and can focus on some key impls.

• Change the Scorer API to expose an iterator instead of extending DocIdSetIterator [LUCENE-6919] #7977
• Stop using SlowImpactsEnum for terms whose docFreq is less than 128. #14017
• Combine all postings enum impls of the default codec into a single class #14033
• Use the same DocIdSetIterator impl for #all, #range and #empty. #14451
• Refactor doc values to expose a DocIdSetIterator instead of extending DocIdSetIterator. #14475
• Make IntervalIterator expose a DocIdSetIterator instead of extending DocIdSetIterator.
• Make Spans expose a DocIdSetIterator instead of extending DocIdSetIterator.
• Can we somehow remove ImpactsDISI, e.g. by evaluating term queries as single-clause disjunctions or conjunctions?