Testing this change on upcoming gpython/py3 with str patched to be ustr
revealed compatibility breakage against several places in standard
library. One example of such a breakage is os.listdir, which after
doing PyObject_CheckBuffer decides to return bytes instead of unicode in
the result:

    https://github.com/python/cpython/blob/v3.11.9-9-g1b0e63c81b5/Modules/posixmodule.c#L4194-L4195

which makes e.g. pytest to fail to work with

    $ gpython -m pytest -vsx
    ...
      File ".../lib/python3.11/pathlib.py", line 370, in _select_from
        if self.match(name):
           ^^^^^^^^^^^^^^^^
    TypeError: cannot use a string pattern on a bytes-like object

This was immediately-seen breakage even without trying to run ERP5 on
top of gpy3. So in general adding buffer interface to ustr is believed to
break too much compatibility with standard unicode on py3 that we
decided against it.

-> Revert adding buffer interface to ustr.

This effectively reverts 8a240b5b (golang_str: Fix ustr to provide
buffer interface, like bstr already does), but leaves added/updated
tests and comments there about why making memoryview(ustr) turned out to
be not a good idea.