Add a test / TODO for a bug in __init__

Summary:
I noticed this in the middle of a refactor, and it's not convenient
to fix until I finish the refactor, but recording a test case.

The correct semantics here aren't 100% clear, but you can see the
runtime behavior [here]( https://codapi.org/embed/?sandbox=python&code=data%3A%3Bbase64%2CSyvKz1UoqSzIzEtXyMwtyC8qUQipLEhNcclMLtFR8MsvCUotLM0sSk3h4krOSSwuVnDRgMtrWnEpgECFlUJmXgmEXQlmK9gqmEL4VVbIhkRn5pXEcnFxFRRl5pVouGhoagIA), the constructor
call allows anything, which means that
 - we almost certainly should allow `z` to be missing
 - we probably should also complain about `y` missing, the generated
   `__init__` does *not* use the default value. I think this bug
	 is caused by incorrect sharing of code between dataclass and typed dict
	 logic, where we shared the code because the operations look the same in
	 the Rust type system, but the actual Python semantics are divergent.

I found this working on a refactor to hide internals of `ClassField`, which
has the side effect of encouraging TypedDict and dataclass to actually own
thier own behavior instead of farming out logic to an overly abstract helper
method, which will naturally make it easy to fix this on top of my refactor.

Reviewed By: ndmitchell

Differential Revision: D70057864

fbshipit-source-id: 52b2aef295b5ee2b3f7b903e04ee32fe51bc2d50

pyre2/pyre2/lib/test/typed_dict.rs

@@ -6,6 +6,7 @@
  */
 
 use crate::testcase;
+use crate::testcase_with_bug;
 
 testcase!(
     test_typed_dict,
@@ -344,3 +345,16 @@ def foo(x: Coord, **kwargs: Unpack[Coord]):
     assert_type(kwargs, Coord)
     "#,
 );
+
+testcase_with_bug!(
+    "TODO(stroxler) We should correctly account for requireness of a field in requiredness of `__init__` args",
+    test_requireness_in_init,
+    r#"
+from typing import TypedDict, NotRequired
+class D(TypedDict):
+     x: int
+     y: int = 5  # E: TypedDict item `y` may not be initialized.
+     z: NotRequired[int]
+D(x=5)  # E: Missing argument `z`
+    "#,
+);