[FEAT] py::vectorize for py::args

I have defined a function signature like [](Type& self, py::args args) -> double where each argument is a scalar (specifically, std::variant<double, int, std::string>), and it would be convenient for my use case if I could wrap this with py::vectorize, where it would do the same as for a static number of arguments (e.g. [](Type& self, double arg1, int arg2) -> double etc.). (although I don't care to vectorize the string)
Do you think this is something that could be supported?

An alternative I thought of (with the help of @henryiii ) would be to first broadcast the arrays and flatten them in pure python and pass them down to an implementation that can handle 1D py::array like the example at the end of https://pybind11.readthedocs.io/en/stable/advanced/pycpp/numpy.html#vectorizing-functions but that comes at the expense of having to hold the broadcasted arguments in memory for some time.

The issue names no source files or tests; start with the py::vectorize documentation and its linked NumPy vectorizing example. Compare the existing fixed-arity behavior with py::args, including broadcasting and mixed scalar types, then establish the supported semantics and tests needed before implementation.