Add the ability to benchmark to create per-field analyzers, like Lucene's PerFieldAnalyzerWrapper [LUCENE-4729]

`AnalyzerFactory`, added to the benchmark module in #5788, is a factory of factories: as part of creating an analyzer, each analysis pipeline component's factory is invoked to create the component instances.

But unlike in Solr, where all analysis pipelines are explicitly specified per-field, benchmark algorithms have no way to specify per-field analyzers - an analyzer is always applied to all fields.

I propose a new task `PerFieldAnalyzerFactoryTask`, which would invoke `PerFieldAnalyzerFactory`: a factory of factories of factories.

(I've left "Wrapper" out of the name - it's already verging on overly long, and I don't think the meaning is compromised.)

I don't have an implementation yet - I wanted to see if people think there's a different/better way to do this.

---
Migrated from [LUCENE-4729](https://issues.apache.org/jira/browse/LUCENE-4729) by Steven Rowe (@sarowe)