Basics of formatting at compile-time based on compile-time API

[alexezeder]

I need to say that this PR does not introduce fully-functioning formatting via {fmt} at compile-time. By fully-functioning I mean that I handled only a few cases (bool, integer, string) of the entire functionality of this library.

This PR based on compile-time API from compile.h. This forces end-users to use FMT_COMPILE macro in cases when formatting should happen at compile-time.

• new compile-time formatting tests added into compile-test test case, available only if C++ standard is greater or equal to C++20
• new FMT_CONSTEXPR20 macro definition added
• new is_constant_evaluated() function in fmt::details namespace added, that invokes std::is_constant_evaluated() if std library has it, otherwise falls back to returning false

What functionality of the library is not available at compile-time with this PR:

• pointers formatting - std::bit_cast is not available yet, so it's not possible yet
• floating-point numbers formatting - looks like it's possible, but takes some time
• format specifications (padding, precision, etc.) - haven't tried, takes some time
• colors, locales, other functionality I'm not aware of - haven't tried, takes some time

[alexezeder]

Since the current CI configuration lacks C++ standards other than C++11 and I was eager to know if it compiles with standards other than C++11, I prepared a special branch to check it on CI, forgot to mention that it has only specific to Linux changes in workflow
https://github.com/alexezeder/fmt/actions/runs/363499597 (looks like g++-10 does not support C++20, so ignore it)

[alexezeder]

Looks like even gcc-10 with 20 standard works now. Not sure why didn't it work for the first time...

[alexezeder]

🤨 I don't know what exactly happened to doc build, but it's was not able to get some dependencies and it's the only one that failed.

[vitaut]

Looks great, just one small suggestion to simplify tests.

[alexezeder]

@vitaut I also remind you that there is a question from me without an answer (#2019 (comment)), this is why I didn't resolve the thread - I fulfilled only one of your points. count_digits_fallback still has template argument.

[vitaut]

Merged, thanks!

[alexezeder]

Since we have the base now, I will try to cover the remaining functionality of the library by following PRs.