fix(parse): allow (to omit) semicolon after _unnamed-declaration_

[JohelEGP]

Follows up on #359.
Resolves #415.
Resolves #432.
Resolves #435.
Resolves #532.
Resolves #545.

Testing summary.

100% tests passed, 0 tests failed out of 866

Total Test time (real) =  50.71 sec

Acknowledgements.

• Tested thanks to https://github.com/modern-cmake/cppfront.
• The commit message uses Conventional Commits 1.0.0.

[JohelEGP]

This change means that

main: () = {
  [[assert: inspect 1 -> bool {
    is 1 = :() -> bool = true;
    is _ = false;
  }]]
}

becomes an ill-formed inspect due to lacking is _.
:() -> bool = true; is _ = false; is the alternative's statement.

[JohelEGP]

@hsutter This one should be easy.

[hsutter]

Thanks!

This change means that

main: () = {
  [[assert: inspect 1 -> bool {
    is 1 = :() -> bool = true;
    is _ = false;
  }]]
}

becomes an ill-formed inspect due to lacking is _. :() -> bool = true; is _ = false; is the alternative's statement.

That's an odd case. Even if the ; ended the alternative's statement, this wouldn't make sense because the alternative result would be a function (that isn't invoked), which shouldn't be what the programmer intended, right? At best it would rely on the implicit function-to-pointer decay to happen to be able to match bool in Cpp1 code).

The likely intended option would be to try to invoke the function:

is 1 = :() -> bool = true;();

• MSVC: works
• GCC (at least 10): gets an error in the generated if constexpr( requires that complains about the attempt to invoke it
• Clang: complains about the lambda in an unevaluated context

Alternatively it could be parenthesized if that's what the programmer really wanted (and unwisely rely on the Cpp1 function-to-pointer decay):

is 1 = (:() -> bool = true);

• MSVC: works
• GCC: works (though GCC apparently-incorrectly warns about control flow reaching the end of a non-void function)
• Clang: complains about the lambda in an unevaluated context