Specialize complex function dispatcher

[hpkfft]

Please consider this PR, which adds a variant of the complex dispatcher that applies when no overloads have kwargs and all overloads have 8 or fewer arguments.

When binding an existing C++ library in order to provide Python support, it's common to have functions that either have default values, or accept nullptr (i.e., None), or have parameters for which noconvert is desirable (especially for nd-array parameters). These are features of C++, as is having only positional arguments.

More than eight parameters is rare. It is arguably a poorly designed API, which would benefit from kwargs in the Python interface (and therefore use the full-featured complex dispatcher).

Given

    auto test_02 = [](int up, int down) -> int { return up - down; };
    m.def("test_02p", (int (*)(int, int)) test_02, nb::arg()=8, nb::arg()=1);
    m.def("test_02nc", (int (*)(int, int)) test_02, nb::arg().noconvert()=8,
                                                    nb::arg().noconvert()=1);

the performance on Python 3.14 is as follows:

test	old	new	ratio
test_02p	27.5ns	22.5ns	1.22
test_02nc	26.9ns	22.5ns	1.20

measured using

python -m timeit -n 100000000 -r 10 -s "import test_functions_ext as t" "t.test_02p(7)"
python -m timeit -n 100000000 -r 10 -s "import test_functions_ext as t" "t.test_02nc(5, 3)"