[FIX] Add handling of composition of standard types

[filipsajdak]

The current implementation does not allow to:

• put variant, pair, tuple inside optional,
• put pair, tuple inside variant,
• put pair inside tuple

Below code will not compile:

p : std::pair<std::string_view, std::optional<std::string>> = ("first", std::nullopt);
std::cout << "p = (p)$\n";

t : std::tuple<double, std::optional<std::pair<std::string_view, int>>, std::optional<std::tuple<int, int, int>>> = (3.14, std::nullopt, std::nullopt);
std::cout << "t = (t)$\n";

vv : std::variant<int, std::string, std::pair<int, double> > = ();
std::cout << "vv = (vv)$\n";

This is just because the order of definitions of to_string specializations matters. When the compiler matches to_string for optional it is not aware of the existence of specializations for variant, pair, or tuple.

This change proposes to add forward declarations to to_string specializations for variant, pair, and tuple.

[JohelEGP]

Reading https://en.cppreference.com/w/cpp/language/dependent_name, seems like qualifying with the calls with cpp2:: binds the lookup for to_string at the context of the template definition. Doesn't this mean that the last overload will never be used?

[hsutter]

Thanks! I've added a followup commit to change the order of the last overload and to add the above test cases to the regression test example.

[filipsajdak]

@JohelEGP Nice catch with the last overload - I completely ignored the fact that it should be matched. Thank you!