[fix] `half_float::half` bad formating

Context

Displaying a half_float::half value throws an error.

#include <fmt/core.h>

#include "aidge/backend/cpu/data/TensorImpl.hpp"
#include "aidge/data/Tensor.hpp"
#include "aige/data/half.hpp"
#include "aidge/utils/Log.hpp"

int main() {
    Aidge::Log::setPrecision(3);
    Aidge::Tensor myTensor{half_float::half{3.5}};
    fmt::print("Value: {}\n", myTensor);
    return 0;
}

throws

terminate called after throwing an instance of 'fmt::v11::format_error'
  what():  invalid format
Aborted (core dumped)

Furthermore, Pybind did not support Numpy fp16, this MR extends Pybind support for NumPy FP16.

Changes

• dev: delegate half_float::half formating to float formatting (which was already partially done)
• fix: Tensor::toString to convert half_float::half element to float before formating
• imp: specialize Tensor constructor from arithmetic value to support half_float::half input
• imp: add support for conversion of numpy.float16, the follwig code now succeed:

a = np.array([5.2616541964], dtype=np.float16)
t = aidge_core.Tensor(a)

• dev: improve unit test on python tensors by testing more numpy dtype.