Wrap a Python io object in an ostream

I claim that it would be worthwhile to allow Python io objects to be wrapped into std::ostreams. Currently, one has to use scoped_ostream_redirect like so:

m.def("foo", [](pybind11::object io) {
    std::ostringstream out;

    pybind11::scoped_ostream_redirect redirect{out, file};

    // rest of code, using `out` as the std::ostream&
});

However, std::ostream can be constructed by a std::streambuf. scoped_ostream_redirect uses a streambuf implementation. Thus, it's possible to directly create a std::ostream from the object:

pybind11::detail::pythonbuf buffer{io};
std::ostream out{&buffer};

What I'm leading up to is this (currently doesn't work):

m.def("foo", [](std::ostream& out) {
    // using `out` as before
});

Which could be used in Python as if it were a regular Python io object:

import io
out = io.StringIO()
m.foo(out)
print(out.getvalue())

Start by reading the scoped_ostream_redirect and detail::pythonbuf implementations mentioned in the issue, then trace how std::ostream& arguments are handled. Done means a Python io.StringIO can be passed to m.foo as std::ostream& and its output is available through getvalue(); no specific files or tests are named.