Fix several redundant bytes/bytearray/memoryview unions

[AlexWaygood]

Unblocks PyCQA/flake8-pyi#201

[github-actions]

According to mypy_primer, this change has no effect on the checked open source code. 🤖🎉

[JelleZijlstra]

I don't feel great about losing precision here. This isn't even specified in PEP 484.

For instance, I just looked at builtins.pyi and the first usage of bytes I found was for ord(): def ord(__c: str | bytes) -> int: .... I looked at the implementation and it allows bytearray, but not memoryview.

Next is the constructor of UnicodeDecodeError. We have it as bytes, but it actually allows any buffer type (so also e.g. array.array).

Then int.__new__. We have one overload that takes bytes (among others) and another that takes bytes | bytearray. Checking the implementation, the first overload allows any buffer type and the second overload actually allows bytes and bytearray, but not memoryview.

I think we're better off reflecting the runtime reality exactly (using _typeshed.ReadableBuffer where appropriate), instead of pretending that these types are all the same.

[AlexWaygood]

Thanks for calling me out. The typing docs seem very bad in this regard. According to this BPO issue, the intended meaning of the entry for ByteString is that "memoryview is sometimes a subtype of ByteString but not all the time". That wasn't obvious at all from my reading of the section, and I don't really understand how it can be put to practical use when using type hints.

Anyway, you're right that this approach is far too sweeping :)