annotationlib.type_repr() returns "None.list.append" for bound built-in methods

Bug report

annotationlib.type_repr() (a public, documented helper, exported in
__all__) returns a string that names a nonexistent module None for bound
built-in methods:

>>> from annotationlib import type_repr
>>> type_repr([].append)
'None.list.append'
>>> type_repr(dict.fromkeys)
'None.dict.fromkeys'
>>> import random; type_repr(random.random)
'None.Random.random'

Bound built-in methods (and C-accelerator functions) are BuiltinFunctionType
with __module__ set to None, so f"{value.__module__}.{value.__qualname__}"
interpolates the literal None. Every other object in this family produces a
resolvable name -- len -> 'len', os.getpid -> 'posix.getpid';
this is the only one that emits a None. prefix. 'None.list.append' is
also an active hazard: a STRING-format consumer that resolves it gets
AttributeError on the literal None.

annotations_to_string() and get_annotations(obj, format=Format.STRING)
propagate it when an annotation value is such a method.

Note for completeness: this is reached when a live method object is passed to
type_repr (its documented purpose), not from natural annotation source
syntax -- def f(x: [].append) is stringified correctly via the AST path.
The defect is the public-API output of type_repr itself.

Fix

Return __qualname__ ('list.append') when __module__ is None,
exactly as already done for the "builtins" module. repr() is not a
usable fallback: it embeds a non-deterministic heap address and is not
re-parseable.

Linked PRs

• gh-152693

Start at the public annotationlib.type_repr() entry point and inspect how it formats objects whose module is None. Verify the documented examples and the annotations_to_string() and get_annotations(..., format=Format.STRING) paths; done means bound built-in methods produce their qualname without a None. prefix and remain resolvable.