_pyio buffered classes raise AttributeError after detach() instead of ValueError #160472

Bug description

After detach(), the pure-Python buffered classes in _pyio raise AttributeError
from the None raw stream, where the C implementations raise ValueError. The two
implementations therefore disagree on the kind of error, not just the wording.

import io, _pyio

for mod in (io, _pyio):
    b = mod.BufferedReader(mod.BytesIO(b"abcdef"))
    b.detach()
    for op, args in [("read", (1,)), ("seek", (0,)), ("truncate", ()), ("flush", ())]:
        try:
            getattr(b, op)(*args)
            result = "no error"
        except Exception as exc:
            result = type(exc).__name__
        print(mod.__name__, op, "->", result)

io read -> ValueError
io seek -> ValueError
io truncate -> ValueError
io flush -> ValueError
_pyio read -> AttributeError
_pyio seek -> AttributeError
_pyio truncate -> AttributeError
_pyio flush -> AttributeError

Same shape for _pyio.BufferedWriter (write, flush, seek, tell),
_pyio.BufferedRandom and _pyio.TextIOWrapper.tell().

Cause

In Lib/_pyio.py, BufferedReader.detach() does

def detach(self):
    if self.raw is None:
        raise ValueError("raw stream already detached")
    self.flush()
    raw = self._raw
    self._raw = None
    return raw

so the class already knows the detached state, but only detach() itself checks
it. Every other method reaches self.raw and fails on the None, producing
AttributeError: 'NoneType' object has no attribute '...'.

Expected behaviour

ValueError, matching the C implementation and matching the documented state of
the object ("After the raw stream has been detached, the buffer is in an unusable
state"). The check C performs is the CHECK_ATTACHED macro in
Modules/_io/bufferedio.c.

Why it matters beyond error text

test_io runs the same test classes against both implementations (C and
Py variants in Lib/test/test_io/), so any test that asserts ValueError for a
detached operation cannot be shared today. A _checkDetached() helper next to the
existing _checkClosed() would let those tests be written once.

Scope

I have not opened a patch, following the guidance in #160467 that the expected
behaviour should be agreed on the issue first. If the ValueError above is the
direction a triager wants, I can prepare the fix plus a test in test_textio.py
and test_bufferedio.py that runs against both implementations.

Found by differential testing _io against _pyio over random operation
sequences.

CPython versions tested on

CPython main branch (3.16.0a1+)

Operating systems tested on

Linux