Update errors in string "Explicit conversions" docs (#4658)

`PyUnicode_DecodeLatin1` requires you to pass in the `error`
parameter. The code as it is in the docs didn't compile.

There is a reference leak in the example
code. `PyUnicode_DecodeLatin1` returns a new reference. Calling
`py::str(PyObject*)` calls `PyObject_Str`, which also returns a new
reference. That reference is managed by the `py::str`
constructor (which correctly steals the reference, using the
`stolen_t` constructor), but the original reference returned by
`PyUnicode_DecodeLatin1` is never decref'd: it never makes it into an
`object`, and it's never manually decremented.

This fixes both of those issues. The code compiles, and I viewed the
sphinx docs locally.

docs/advanced/cast/strings.rst

@@ -101,8 +101,11 @@ conversion has the same overhead as implicit conversion.
     m.def("str_output",
         []() {
             std::string s = "Send your r\xe9sum\xe9 to Alice in HR"; // Latin-1
-            py::str py_s = PyUnicode_DecodeLatin1(s.data(), s.length());
+            py::handle py_s = PyUnicode_DecodeLatin1(s.data(), s.length(), nullptr);
-            return py_s;
+            if (!py_s) {
+                throw py::error_already_set();
+            }
+            return py::reinterpret_steal<py::str>(py_s);
         }
     );
 
@@ -113,7 +116,8 @@ conversion has the same overhead as implicit conversion.
 
 The `Python C API
 <https://docs.python.org/3/c-api/unicode.html#built-in-codecs>`_ provides
-several built-in codecs.
+several built-in codecs. Note that these all return *new* references, so
+use :cpp:func:`reinterpret_steal` when converting them to a :cpp:class:`str`.
 
 
 One could also use a third party encoding library such as libiconv to transcode