Allow use of `Required` and `NotRequired` to make an existing typed dict total or optional

I think the ability of indicating that a typed dict requires all keys or that all are optional would be very useful.

For example to allow defining default values for a particular dict or to return a dict that has been completely assigned from a function

A simple example

class A(TypedDict):
    a: int
    b: int

ADefault = {'a': 1, 'b': 42}

def optionalA(v: NotRequired[A]) -> A:
    return ADefault | v

optionalA({'a': 11}) # ok since all keys of NotRequired[A] are optional

class X(TypedDict, total=False):
    x: str
    y: str

XDefault: X = {"x": "xv", "y": "yv"}

def withXDefault(v: X) -> Required[X]:
    return XDefault | v

withXDefault({"x": "foo"})["y"]  # ok since all keys of Required[X] are required

This idea is inspired by the analogous types in typescript Required and Partial that are in general very convenient.

I think it makes sense having the same behaviour of the typescript counterparts:

• NotRequired[X] is a no op, since X is defined as total=False. Same for Required[A].
• Required/NotRequired would include all keys of the typed dict and all its superclasses. This differs from how total works. Example:

  class TD1(TypedDict):
      a: int
  
  class TD2(TypedDict, total=False):
      b: int
  
  TDR = Required[TD2]  # b becomes required
  TDNR = NotRequired[TD2]  # a becomes not required
  

• Required/NotRequired used inline in the TypedDict are not taken into consideration, so the end result is the same independently of how the typed dict is defined

Start by reviewing the proposed TypedDict, Required, and NotRequired semantics and the examples in the issue, including inheritance and inline annotations. Define the behavior and identify the relevant typing specification and conformance tests; done means the requested transformations are specified and validated consistently.