Half float 16 descriptor

[swillisart]

Hello!
I am a python developer working on enabling support for half float 16 types in numpy array templates.
I'm planning on creating a PR for this once it's ready but I'm hitting a roadblock due to my limited knowledge of C++.

I have enabled the format character code e and the half constant in numpy.h, but it doesn't seem to automatically detect the half type from Imath (https://github.com/AcademySoftwareFoundation/Imath).

How do I make pybind aware that Imath::half is supposed to be interpreted as numpy half dtype? Is it something that needs to be declared on the source?

Since float and double are native types I'm not sure how to modify the templates to automatically map the unknown type.
I would like to avoid pasting this boilerplate in bindings. (From #1776)

#include <half.h>
namespace py = pybind11;
namespace pybind11 {
namespace detail {
    template<> struct npy_format_descriptor<half> {
        static py::dtype dtype() {
            handle ptr = npy_api::get().PyArray_DescrFromType_(npy_api::NPY_HALF_) ;
            return reinterpret_borrow<py::dtype>(ptr);
        }
        static constexpr auto name = _("float16");
    };
}}

Thank you for reading!