Specialize __setitem__ dunder method for STORE_SUBSCR

This is precursuor for JIT improvements but also results in
standard improvements due to specializing the bytecode

Author: johng

Include/internal/pycore_code.h

@@ -131,6 +131,7 @@ typedef struct {
 
 typedef struct {
     _Py_BackoffCounter counter;
+    uint16_t version[2];  /* gh-143732: container type version for STORE_SUBSCR_PY_DUNDER */
 } _PyStoreSubscrCache;
 
 #define INLINE_CACHE_ENTRIES_STORE_SUBSCR CACHE_ENTRIES(_PyStoreSubscrCache)

Include/internal/pycore_opcode_metadata.h

@@ -410,6 +410,8 @@ def wrap_func_w_kwargs():
                LOAD_SMALL_INT           0
                CALL                     1
                STORE_SUBSCR
+               CACHE
+               CACHE
                LOAD_COMMON_CONSTANT     7 (None)
                RETURN_VALUE
 """

Include/internal/pycore_uop_ids.h

@@ -691,6 +691,36 @@ def write(items):
         opname = "BINARY_OP_SUBSCR_GETITEM"
         self.assert_races_do_not_crash(opname, get_items, read, write)
 
+    @requires_specialization
+    def test_store_subscr_py_dunder(self):
+        def get_items():
+            class C:
+                __setitem__ = lambda self, item, value: None
+
+            items = []
+            for _ in range(self.ITEMS):
+                item = C()
+                items.append(item)
+            return items
+
+        def read(items):
+            for item in items:
+                try:
+                    item[None] = None
+                except TypeError:
+                    pass
+
+        def write(items):
+            for item in items:
+                try:
+                    del item.__setitem__
+                except AttributeError:
+                    pass
+                type(item).__setitem__ = lambda self, item, value: None
+
+        opname = "STORE_SUBSCR_PY_DUNDER"
+        self.assert_races_do_not_crash(opname, get_items, read, write)
+
     @requires_specialization
     def test_binary_subscr_list_int(self):
         def get_items():

Include/internal/pycore_uop_metadata.h

@@ -0,0 +1,4 @@
+Specialize :opcode:`STORE_SUBSCR` for container objects whose type defines
+``__setitem__`` as a simple Python function.  The new ``STORE_SUBSCR_PY_DUNDER``
+specialization guards on the container's type version and calls the dunder
+directly, skipping the generic ``PyObject_SetItem`` slot dispatch.