Check skeleton RID when using cached AABB

[ecmjohnson]

Resolves #73521 by ensuring that the RID of the provided skeleton matches that used to compute the AABB of the mesh.

This comment presents the root cause of the issue.

[clayjohn]

This makes sense, but I am a bit worried about the performance result. With this change we could end up recomputing the AABB multiple times per frame as it is never cached for a skeleton/mesh pair.

We should almost be caching on the MeshInstance level so we can guarantee that each mesh has a single skeleton. After all, this code is getting called in _update_instance_aabb()

[clayjohn]

Actually, thinking through it some more, this is probably fine for now. We only call this function when the Instance knows it needs to update the AABB because something has changed. So the assumption is that, if this is called, it will actually do an update.

Since the AABB is cached by the Instance already, the extra level of caching here doesn't help much.

[akien-mga]

Thanks!