The fix made for https://bugs.openjdk.java.net/browse/JDK-8074346 is proving to a bit problematic
due to originally unforeseen side effect.

More permissively attributing expecting either a package name or a type name - the central idea behind
the fix for JDK-8074346 - while conceptually is the correct thing to do - implementation wise exhibits
some problems:

As an inspection of com.sun.tools.javac.comp.Resolve#findIdent shows, this results in an existing
type being preferred over a package - but if no existing type is found, a package symbol is created
and returned without checking if such a package symbol exists in the first place !!!

So the attempt to attribute SomeException in (the test case for
https://bugs.openjdk.java.net/browse/JDK-8130401)

public class ThrowsException {
    public void method() throws @Annot SomeException {}
}

returns a valid PackageSymbol for SomeException ! and we then conclude
that a type annotation cannot be applied to a package component and so
issue an error about how TAs cannot be applied to scoping constructs.

This bogus error results in the compilation unit as being containing
a non-resolution error - thus resulting in annotation processing being
skipped.

I propose to revert the fix for JDK-8074346 so that a comprehensive fix for
https://bugs.openjdk.java.net/browse/JDK-8074346 and
https://bugs.openjdk.java.net/browse/JDK-8130401 and (possibly)
https://bugs.openjdk.java.net/browse/JDK-8130386 can be explored.