The plan is to hide the javadoc into sealed interfaces, many per base type, so that the main class is less bloated when looking at its source code.

It works, but undermines many of the javadoc self tests we've been using to avoid common pitfalls while documenting RxJava.

Since the self tests are hand-crafted string splicers, we might just augment them to look up the method in the sealed interface and inline its javadoc before the check happens.

A better alternative would be to use a proper Java parser which then we could navigate the sources and docs more easily.

Could also replace many other hand crafted checks.