* Allow disabling floating point support

Add FMT_USE_FLOAT, FMT_USE_DOUBLE and FMT_USE_LONG_DOUBLE to allow a
user of the library to configure the float types they want to allow.
This is specially useful in embedded environements where code size is
important.

* Avoid conditional macros to disable float support

* Add is_supported_floating_point constexpr function

* Fix empty-body warning

* Use std::char_traits::length for ansi_color_escape::begin

-Fixes issue #1594 https://github.com/fmtlib/fmt/issues/1594

FMT_DEPRECATED is now defined as FMT_HAS_CPP14_ATTRIBUTE(deprecated), as this attribute was introduced in C++14.

FMT_FALLTHROUGH is now defined as FMT_HAS_CPP17_ATTRIBUTE(fallthrough), as this attribute was introduced in C++17.

FMT_MAYBE_UNUSED is defined as FMT_HAS_CPP17_ATTRIBUTE(maybe_unused), as this attribute was introduced in C++17.

FMT_MAYBE_UNUSED has been applied to fix a couple of -Wunused-member-function warnings from clang.

* Workaround broken [[deprecated]] in PGI compiler
  - similar to Intel and NVCC, add workaround for PGI compiler

FMT_STRING_IMPL has an internal helper named FMT_STRING, however FMT_STRING is also the name of the macro that invokes FMT_STRING_IMPL.

Renaming this helper avoids the appearance of a recursive macro.

grisu_count_digits is only used by grisu_gen_digits, which assigns the unsigned result to a (signed) int.

Although grisu_count_digits always returns a positive integer this keeps its return type in sync with the type its result is assigned to.