The option to plug meta-data generator tool as a service has been considered. But it was not activated it because we did not know how the users will react if pure javac compilation with OpenJPA library in compiler classpath starts generating a bunch of MyEntity_.java files.
Our assumption is only a handful of brave users will start experimenting with new Criteria API and a subset of them will use the typesafe version that requires these automatically generated canonical classes. That is why, the automatic generation is not activated as service.
But one is always wrong in making any assumption on usage pattern
About location of the generated sources:
Our first preference was to generate the new MyEnity_.java files in the same directory as in MyEntity.java, by default.
But it is not obvious how to determine the directory location of a *.java files in the context an AnnotationProcessor is invoked. Nor the ways to access the javac options such as -sourcepath etc.
After a discussion in this regard  and some basic experiments, I concluded that the location of MyEntity.java can only be determined with a specialized API in Sun JDK. Based on that observation, the default location of the generated files are current directory (for non-Sun JDK) and the directory of the original source for Sun JDK.