Fix fmt::formatted_size() with compiled format (FMT_COMPILE) as argument

[alexezeder]

Resolves #2141.

All useful info can be found in the issue, including code examples.

One small detail for this PR - fmt::formatted_size() in compile.h becomes templatized too to eliminate one unclear case - https://godbolt.org/z/Yxnccj. In few words, the implementation of fmt::formatted_size() for runtime API (non-compiled format string) changes depending on whether we include fmt/compile.h or not. You can see it in assembly for the formatted_size, it uses counting_iterator, which is used in the overload from compile.h:

fmt/include/fmt/compile.h

Lines 837 to 840 in 835b910

	template <typename CompiledFormat, typename... Args>
	size_t formatted_size(const CompiledFormat& cf, const Args&... args) {
	return format_to(detail::counting_iterator(), cf, args...).count();
	}

but not in the overload from core.h:

fmt/include/fmt/core.h

Lines 1864 to 1870 in 835b910

	template <typename... Args>
	inline size_t formatted_size(string_view format_str, Args&&... args) {
	const auto& vargs = fmt::make_args_checked<Args...>(format_str, args...);
	detail::counting_buffer<> buf;
	detail::vformat_to(buf, format_str, vargs);
	return buf.count();
	}

Maybe it's okay, but IMHO in this case, that behavior should be documented, right?

[vitaut]

the implementation of fmt::formatted_size() for runtime API (non-compiled format string) changes depending on whether we include fmt/compile.h or not.

Good catch, formatted_size in fmt/compile.h was previously underconstrained.