Ensure that unannotated known class types are singletons

[snazy]

In particular, this commit prevents "changing" an unbounded wildcard ? to ? extends java.lang.Object. This was caused by a new ClassType instance being created for java.lang.Object, instead of reusing the singleton instance.

[Ladicek]

I've noticed this commit in the previous PR and didn't pay much attention, but I'd like to ask here: what issue does this fix? The PR looks reasonable, but still.

[snazy]

I've seen the same ? generic being serialized as a simple ? (annotation-instance referring to ClassType.OBJECT_TYPE, which is then "simplified" as ?) and serialized as ? extends java.lang.Object (annotation instance to a different ClassType for java.lang.Object).

So I went ahead and added the "forceful" indirection via ClassType.create() to ensure that the "proper singletons" are being used everywhere.

[snazy]

This seemed much easier than an instance-equals at the call sites.

[Ladicek]

Ah so accidental discovery. OK, I think this makes sense; ? extends java.lang.Object with no annotation on the bound is the same as ?.

[snazy]

Or see it as a micro-heap-pressure-optimization ;)