Add faster versions of various runtime-checkable protocols

[AlexWaygood]

This is one way of resolving the disagreement in #143. The other way would be to just revert #141. No strong opinion either way :)

[AlexWaygood]

(If this PR is accepted, I'll update #143 to talk about the new additions to typing_extensions in the README.)

[srittau]

Looking back at #141, I'm not sure how relevant the speedups are and whether it's worth complicating typing_extensions for it. But I don't feel strongly about either. So, -0 and +0 on reverting #141.

[AlexWaygood]

I'm not sure how relevant the speedups are

I'd argue the speedups are quite relevant. The optimisations we've introduced into Python 3.12 were designed to address a longstanding user report where _ProtocolMeta.__instancecheck__ was revealed to be a significant bottleneck in their code: python/cpython#74690. Elsewhere, the slowness of _ProtocolMeta.__instancecheck__ has led the beartype library to completely reimplement _ProtocolMeta in order to address the performance issues: https://github.com/beartype/beartype/blob/main/beartype/typing/_typingpep544.py.

I think it's fine to decide that we don't want to bloat the API of typing_extensions, or that it's not our job to backport things just to make them faster. But I'll defend the relevance of the optimisations here to our users; I think the speedup addresses a real problem that's been a common pain point for users of typing.

[JelleZijlstra]

Thinking about it more I'm OK with merging this. Performance improvements are new features, and typing-extensions is about backporting new features.