[BUG] NumPy fixed width integer types don't cast correctly

Issue description

NumPy fixed width integer types don't cast correctly.

Reproducible example code

#include <cstdint>
#include "pybind11/pybind11.h"
namespace pb11 = pybind11;

template<typename T>
T x2(T v)
{
    return v + v;
}

PYBIND11_MODULE(OverloadTest, m)
{
    m.def("x2", &x2<std::uint8_t>);
    m.def("x2", &x2<std::uint16_t>);
    m.def("x2", &x2<std::uint32_t>);
    m.def("x2", &x2<std::uint64_t>);
}

import numpy as np
import OverloadTest as ot

print(ot.x2(np.uint64(200)))

which outputs

>> 144

So it seems like the np.uint64 is getting cast to the std::uint8_t type instead of the std::uint64_t type.

Start by compiling and running the reproducible C++ OverloadTest module with the Python np.uint64 example. Inspect pybind11's overload and fixed-width integer casting behavior, then verify that the uint8_t, uint16_t, uint32_t, and uint64_t overloads select the matching NumPy integer types and return the expected results.