[FEAT] format_descriptor of pair<float,float>

The following code, using std::array, is working fine

#include <pybind11/pybind11.h>
#include <pybind11/stl.h>
#include <pybind11/stl_bind.h>
#include <pybind11/numpy.h>

namespace py=pybind11;
typedef std::array<double,2> P;
typedef std::vector<P> V;
PYBIND11_MAKE_OPAQUE(V);
PYBIND11_MODULE(u, m) {
  py::bind_vector<V>(m, "VectorPair", py::buffer_protocol());
  m.def("fff",[](){
      V a(4000, {1,2});
      return py::array(py::cast(std::move(a)));
      });
}

However, if I replace the array with

typedef std::pair<double,double> P;

it fails to compile

static_assert(vector_has_data_and_format::value, "There is not an appropriate format descriptor for this vector");

and I need to add for instance

namespace pybind11 { template<>struct format_descriptor<P>:format_descriptor<double[2]>{}; }

to let it work again.

I was wondering if it would make sense for pybind11 to provide such specializations for std::pair and std::tuple when all the types are the same, like it already does for std::array. One drawback I can think of is that when the types differ, say std::pair<int,float>, we probably want something like PYBIND11_NUMPY_DTYPE(P, first, second); instead (with some standard libraries we may also need to fake namespace std { template<>struct is_trivially_copyable<P>:true_type{}; }), and that would make the case of equal types an exception. I guess I would also be ok with getting a structured array, I can always use view to turn it into the array I want...

#1495 is the closest issue I found in the list, but different.

Start by tracing the format_descriptor specializations and the vector_has_data_and_format assertion involved in py::array conversion, then compare the existing std::array handling with std::pair and std::tuple. Done means the supported same-type pair and tuple cases compile and behave as intended, while differing element types remain compatible with the documented structured-dtype approach.