WikitextContentHandler should support using Parsoid for generating rendered HTML, controlled by a flag in ParserOptions. This way, we will be able to remove most of the special case code for generating parsoid output. It will also allow parsoid to benefit from the infrastructure in place for other kinds of rendering, such as stampede protection using PoolCounter and handling of multiple slots.
NOTE: For some time, we will be generating wikitext renderings using the old parser as well as parsoid, and we will want to store both in the parser cache. But not in the same instance, and possibly in a different backend. To allow this, ParserOutputAccess must allow routing output to difference ParserCache instances based on ParserOptions (and content model?).
NOTE: Once this is done, parsoid output becomes available via ParserOutputAccess. At that point, ParsoidOutputAccess class should be removed.