ENH: Add aliases for commonly used dtype-like objects

[BvB93]

This PR adds (invariant) aliases for commonly used dtype-like objects, similar to what #18050 has done for array-likes.

Note that the aliases introduced herein are currently unused (beyond their introduction).
This will change once we start adding dtype-support to the ufuncs, something that's on
the to-do list for after the ufunc-based ndarray magic methods have been wrapped up (xref #18128 and #18228).

Secondly, it performs a small follow-up on #13578, i.e. objects with the .dtype attribute are
now only considered dtype-like if aforementioned attribute returns a np.dtype instance.

Examples

Example usage of the new _DTypeLike<x> aliases in annotating a simplified version of np.bitwise_or.

from __future__ import annotations
from typing import Any, overload

import numpy as np
import numpy.typing as npt

@overload
def bitwise_or(
    __x1: npt._ArrayLikeBool_co, 
    __x2: npt._ArrayLikeBool_co,
    *, dtype: npt._DTypeLikeBool | None = ...,
) -> np.ndarray[Any, np.dtype[np.bool_]]: ...
@overload
def bitwise_or(
    __x1: npt._ArrayLikeUInt_co, 
    __x2: npt._ArrayLikeUInt_co, 
    *, dtype: npt._DTypeLikeUInt | None = ...,
) -> np.ndarray[Any, np.dtype[np.unsignedinteger[Any]]: ...
@overload
def bitwise_or(
    __x1: npt._ArrayLikeInt_co, 
    __x2: npt._ArrayLikeInt_co, 
    *, dtype: npt._DTypeLikeInt | None = ...,
) -> np.ndarray[Any, np.dtype[np.signedinteger[Any]]: ...

[BvB93]

A typical example of the newly introduced _DTypeLike<X> unions and their components:

• A np.generic subclass.
• A builtin counterpart to aforementioned np.generic subclass.
• A np.dtype instance.
• An object with the dtype attribute.
• Various string literals with relevant character codes ("f", "float", "f4", etc.).

[rgommers]

Thanks @BvB93, this looks good to me modulo the one typo.

[rgommers]

Merged, thanks!