Test and fix for second grandchild bug.

[CrazyCasta]

Bug is, second grandchild of a given child will get cast to the child
rather then the grandchild.

[CrazyCasta]

Just FYI, I haven't been able to run runtest.sh so I don't have coverage information for this yet. I'll be happy to run coverage as soon as someone can explain what's wrong (from #42).

[CrazyCasta]

Okay and the build failed because it timed out downloading some stuff so if someone could show me how to restart the build, that would be much appreciated too.

[CrazyCasta]

I'm at an evaluation point now. It all looks good to me, but I want to point out that the behavior of select_subclasses is changed in Django >= 1.6.0 when a grandchild class is specified. Previously both the child and grandchild would be automatically subclassed when only the grandchild was specified. Now only the grandchild is subclassed. I would argue that this is better because it allows the programmer to more specifically choose what classes to subclass (which is why they were specifying them specifically after all).

Specifically the change in the test is to the test_select_specific_grandchildren function:

-                    self.child1,
+                    InheritanceManagerTestParent(pk=self.child1.pk), 

[carljm]

This all looks great to me, and I agree with that behavior change. I'm not concerned about backwards-compatibility since so few people are likely to be using model-utils with grandchildren at all at this point. Merging!