Duplicate keys cause earlier TypedDict values to be skipped during type checking

[Hmanbo]

Bug Report

When a dictionary literal used to construct a TypedDict contains duplicate
keys, mypy only type-checks the final value for that key. Earlier value
expressions are skipped entirely.

This can hide real errors because Python still evaluates every value expression
in the dictionary literal, including values that are later overwritten by a
duplicate key.

To Reproduce

from typing import TypedDict


class Payload(TypedDict):
    value: int


def parse(value: str) -> int:
    return len(value)


payload = Payload({"value": parse(123), "value": 0})

$ mypy --show-error-codes --no-incremental example.py
Success: no issues found in 1 source file

Running the program demonstrates that the skipped expression is evaluated:

TypeError: object of type 'int' has no len()

Expected Behavior

Mypy should type-check every value expression in the dictionary literal, even
if a later duplicate key determines the final value of the TypedDict field.
In this example, it should report an error similar to:

error: Argument 1 to "parse" has incompatible type "int"; expected "str"  [arg-type]

The same expression is correctly rejected when the target is a regular
dict[str, int]. The issue also occurs with a contextually typed literal:

payload: Payload = {"value": parse(123), "value": 0}

Actual Behavior

Mypy reports no issues because only the last value associated with the
duplicate key is checked.

Additional Context

The current implementation replaces the previously collected expression
without type-checking it and has a TODO to type-check all duplicate-key values:

mypy/mypy/checkexpr.py

Lines 839 to 842 in 6eb141f

	# A directly present key unconditionally shadows all previously found
	# values from ** items.
	# TODO: for duplicate keys, type-check all values.
	result[literal_value] = [item_arg]

PR #15425 also mentioned repeated TypedDict keys as a known limitation, but I
could not find an existing issue that tracks it:

#15425

Environment Used to Reproduce

• Mypy versions used:
  • mypy 2.3.0 (compiled: yes)
  • mypy 2.4.0+dev.6eb141f03a8b43c274f15cc4978d617a71d8f37e (compiled: no)
• Mypy command-line flags: --show-error-codes --no-incremental
• Mypy configuration options: none
• Python version used: 3.12.8

[Hmanbo]

I'd like to work on this.

My plan is to preserve all explicit values associated with a duplicate key
for type checking, while keeping the final value first for TypedDict generic
type inference. I'll also add regression tests for both TypedDict constructor
calls and contextually typed dictionary literals.

Does this approach sound reasonable?