Incorrect dtype-ctypes static typing

[mixilchenko]

I have some c++ - python compatible features and use ctypes lib for creating dtypes dynamically.
np.dtype typing is incorrect when ctypes type is argument.

Reproducing code example:

# myproblem.py
import ctypes
from typing import TYPE_CHECKING

import numpy as np


dtype_int = np.dtype(ctypes.c_int)
dtype_size_t = np.dtype(ctypes.c_size_t)
print(repr(dtype_int))  # prints `dtype('int32')>` as expected on my machine
print(repr(dtype_size_t))  # prints `dtype('uint64')` as expected on my machine
if TYPE_CHECKING:
    reveal_type(dtype_int)  # something like `numpy.dtype[numpy.signedinteger[numpy.typing._32Bit]]` is expected
    reveal_type(dtype_size_t)  # something like `numpy.dtype[numpy.unsignedinteger[numpy.typing._64Bit]]` is expected

Mypy error message:

$ mypy test.py
myproblem.py:8: error: Value of type variable "_DTypeScalar_co" of "dtype" cannot be "c_int"
myproblem.py:9: error: Value of type variable "_DTypeScalar_co" of "dtype" cannot be "c_size_t"
myproblem.py:13: note: Revealed type is "numpy.dtype[ctypes.c_int*]"
myproblem.py:14: note: Revealed type is "numpy.dtype[ctypes.c_size_t*]"
Found 2 errors in 1 file (checked 1 source file)

NumPy/Python version information:

>>> import sys; from importlib.metadata import version; print(sys.version, version('mypy'), version('numpy'), sep='\n')
3.9.1 (default, Jul  7 2021, 22:39:47)
[Clang 12.0.5 (clang-1205.0.22.9)]
0.910
1.21.0

[BvB93]

Ah, I though we had a catchall overload for miscellaneous typeobjects, but it seems to have been removed at some point.

In any case, #19468 should add the necasary overloads for the ctypes types.