The reason we proposed switching from the pybind11 to python is that pybind11 conflicts with the main pybind11 namespace - even if it's aliased to py, it's still available as pybind11. And while e.g. lsst::afw::geom::pybind11::Foo will be selected in preference to ::pybind11::Foo if you just say pybind11::Foo within lsst::afw::geom, I think it's still confusing.
I don't have strong feelings about wrapFoo vs. declareFoo. If there's broad consensus, I don't think it'd be hard to change, though.
I've added the requested examples, added a rule about when to wrap enums as integer attributes (an excellent suggestion), and clarified that the division rule applies to in-place operators.