Add input adapter tests

[nlohmann]

This PR adds tests for the new input adapter implementation from #2145.

[nlohmann]

@FrancoisChabot I added some tests to check all possible character types. Strangely, the UTF-16 and UTF-32 tests fail on MSVC 2015, see https://ci.appveyor.com/project/nlohmann/json/builds/33373571. Any idea on this?

[coveralls]

Coverage remained the same at 100.0% when pulling ae04426 on input_adapter_tests into 28048d8 on develop.

[FrancoisChabot]

I think I know what this is. I ran into a very similar MSVC bug years ago: Older MSVC just seem to sometimes randomly choke when invoking a constexpr function during SFINAE tests.

https://stackoverflow.com/questions/46144103/enable-if-not-working-in-visual-studio-when-using-a-constexpr-function-as-argume

in #2145, I spun off an enable_if<> test as the is_iterator_of_multibyte() function, because astyle was failing to format the inlined version correctly, which created a suspiciously similar setup to the original bug. It's not the exact same compiler version, but it's too similar to be a conicidence.

This would cause the input adapter to treat all data as UTF-8. This explains:

1. The truncation warnings as the int16_t and int32_t gets squished into a int8_t
2. The failing tests. Since the wide_string_input_helper is not executed at all.

So a fix would be to ditch the constexpr function and use an old-school type-based test instead.

I don't have time to draft a full PR for this right now, but the following should fix the issue if my theory is correct:

template<typename T>
struct is_iterator_of_multibyte {
   using value_type = typename std::iterator_traits<T>::value_type;
   enum {
    value = sizeof(value_type) > 1
   };
};

template<typename IteratorType>
struct iterator_input_adapter_factory<IteratorType, enable_if_t<is_iterator_of_multibyte<IteratorType>::value>>

[nlohmann]

Thanks! I'll give it a try.

[nlohmann]

Looks good, thank you!