Current plan is to first do T297840: Populate ParserCache with Parsoid output as a post-edit hook for pages in the Talk namespace, and then follow that up by using Parsoid to render pages synchronously in the read request in the place of the legacy parser. This ensures that the added latency from Parsoid won't be visible in 99% of the cases, and we can watch ParserCache utilization carefully *before* actually enabling read views.