Fix annotated assignment expressions being treated as unpacking targets

[dchaudhari7177]

Summary

Fixes #557. visit_NamedExpr wrapped its single target in an extra list, emitting

check_variable_assignment(otherfunc(), [[('x', int)]], memo)

where the plain-assignment path emits [('x', int)] for a single name. check_variable_assignment reads a list group as a tuple-unpacking target and a tuple group as a single target, so an annotated walrus took the unpacking branch and consumed the assigned value with list(value).

Two wrong outcomes, both only when the name carries an annotation:

assigned value	before	after
5	TypeError: 'int' object is not iterable	5
None	TypeError: 'NoneType' object is not iterable	None
"abc"	no error, x becomes ['a', 'b', 'c']	"abc"
"not an int" (wrong type)	TypeCheckError	TypeCheckError (unchanged)

The third row is the dangerous one — no exception, just a silently corrupted variable.

A walrus target is always a single Name, so it now emits a bare tuple like the assignment path does. Type checking itself is unaffected: a genuinely wrong value still raises TypeCheckError.

Tests

New TestAssignmentExpression in test_typechecked.py covers the behaviour rather than the emitted source: non-iterable value, iterable value not consumed, an annotated argument reused in a walrus, an iterator not drained, and a wrong-typed value still raising.

Reverting only src/ fails 4 of the 5 (the 5th passes either way, since it asserts the error that already worked):

FAILED TestAssignmentExpression::test_non_iterable_value
FAILED TestAssignmentExpression::test_value_is_not_consumed
FAILED TestAssignmentExpression::test_annotated_argument
FAILED TestAssignmentExpression::test_iterator_value_is_not_consumed

test_transformer.py::TestAssign::test_assignment_expr and test_assignment_expr_annotated_argument asserted the old [[('x', int)]] output, i.e. they encoded the defect, so they now expect the single-target form. Worth noting those two golden tests passed throughout the bug's lifetime — pinning emitted source didn't catch it, which is why the new tests assert runtime behaviour instead.

Verification

• Full suite (--ignore=tests/mypy) → 516 passed, 7 skipped, 9 xfailed
• ruff check src tests → all checks passed; ruff format --check → 31 files already formatted
• Added an UNRELEASED entry to docs/versionhistory.rst

tests/mypy is excluded above: both tests there fail with FileNotFoundError on a clean tree too (no mypy binary in my environment), unrelated to this change.

[coveralls]

coverage: 94.847%. remained the same — dchaudhari7177:fix/annotated-walrus-instrumentation into agronholm:master

[dchaudhari7177]

Removed, thanks — the reasoning lives in the PR description instead.

For reference, the emitted call goes from [[('x', int)]] to [('x', int)]: the inner list marked the group as an unpacking target, so check_variable_assignment consumed the value with list().

pytest tests/test_transformer.py still passes (84 passed, 1 skipped). The 2 failures in the full run are tests/mypy/ looking for a mypy binary that isn't installed here.