Mark LRU cache wrappers as method descriptors (#9072)

Match the Python 3.14 wrapper type flag and enable the existing guarded method-call fast path. Keep the partial vectorcall changes separate.

Assisted-by: OpenAI Codex:model identifier unavailable

crates/vm/src/stdlib/_functools.rs

@@ -842,7 +842,7 @@ mod _functools {
 
     #[pyclass(
         with(Constructor, Callable, GetDescriptor),
-        flags(HAS_DICT, HAS_WEAKREF)
+        flags(HAS_DICT, HAS_WEAKREF, METHOD_DESCRIPTOR)
     )]
     impl Py<PyLruCacheWrapper> {
         #[pymethod]

extra_tests/snippets/stdlib_functools.py

@@ -1,9 +1,13 @@
 import gc
 import weakref
+from _functools import _lru_cache_wrapper
 from functools import cache, cmp_to_key, lru_cache, partial, reduce
 
 from testutils import assert_raises
 
+# Cached wrappers support method-descriptor lookup and binding.
+assert _lru_cache_wrapper.__flags__ & (1 << 17)
+
 
 class Squares:
     def __init__(self, max):