Add some miscellaneous tests

[brianschubert]

Add tests to exercise some previously unexercised lines:

• _SpecialGenericAlias.__setattr__() for non-dunder, non-"allowed" attributes (src)
• Wrong number of type arguments given to a _SpecialGenericAlias from without defaults (src)
• Attempting to specialize a _SpecialGenericAlias with 0 type parameters (src)
• Setting <thing>.__module__ on Python implementations that don't supply sys._getframemodulename or sys._getframe (src)
• Calling get_type_hints() with types.GenericAlias (src)
• Exception raised by __init_subclass__() of ParamSpec (src)
• The methods __or__() and __ror__() of _SpecialForm (src)
• The methods __repr__ and __reduce__ of NoExtraItems (test case copied from NoDefaultTests) (src)

[AlexWaygood]

nice, thank you!

[brianschubert]

Thanks for the review! I pushed a few changes. re: making ParamSpec.__init_subclass__ raise in the 3.9 implementation, would this be worth a changelog mention? For reference, the 3.10+ implementation started raising in #165, which didn't include a changelog entry

[Daraan]

making ParamSpec.init_subclass raise in the 3.9 implementation, would this be worth a changelog mention? For reference, the 3.10+ implementation started raising in #165, which didn't include a changelog entry.

Imo, it's worth noting. That it is now consistent with the other versions.

[AlexWaygood]

Thanks for the review! I pushed a few changes. re: making ParamSpec.__init_subclass__ raise in the 3.9 implementation, would this be worth a changelog mention? For reference, the 3.10+ implementation started raising in #165, which didn't include a changelog entry

Probably worth it, just to be safe? I feel like #165 probably should have had a changelog mention too if it changed behaviour 😄

[AlexWaygood]

Looks great! I think there's a merge conflict