Late resolution of reflected types

Moved here from SourceForge RFE 1592498:

Currently when the class object of a type is requested we resolve the type. This avoids checking whether the type is resolved in reflection code. This could mean
we're resolving things earlier than we should (potential bug) and also resolving types we don't need to resolve (potential slow down). We should review the
reflection and type code with a view to making resolution follow whatever the normal JVM convention is.