Unable to get detailed numpy type info into C++ code from dtype

Created by: chuckatkins

The pybind11::dtype object allows access to low-level type information via the kind() method, which maps to the numpy.dtype.kind python property. This is only a coarse level of type information and is missing sizing. For instance, pybind11::dtype::of<float>().kind() is equal to pybind11::dtype::of<double>::kind() because they are both floating point types. The numpy.dtype class exposes more detailed information via the type property but that's not exposed in pybind11, thus leaving only the coarser level type-information.