Add a `bone_collection.bones_recursive` property that returns the set of
bones assigned to that bone collection or any of its child collections.
This property is implemented in Python, as that made it considerably
simpler to create the set semantics.
Use `bpy.ops.wm.read_homefile(use_factory_startup=True)` to load the
default 'homefile'. Otherwise Blender will load the saved-as-default file,
which can be quite different from the default startup file.
Add a read-only property `bone_collection.parent` to RNA that returns
the parent bone collection.
This performs two scans of the array (one to find the bone collection's
index, and the other to find the parent index). This might look bad, but
as long as `Object.children` still loops, in Python, over all of
`bpy.data.objects`, this should also be acceptable.
Add a Python unit test that covers some of the armature/bone collections.
Some of these tests cover the same functionality as the C++ tests, albeit
via the Python/RNA API.
Another test was added to check that joining armatures works as expected.
