The future of fpclassify.hpp

I'm opening this as a place for discussion really.

We currently have boost::math::fpclassify/isnan/isinf/isnormal etc which began life as workarounds for C++03 compilers, but they have another role: for IEEE conforming types they work even in the presence of -ffast-math.

Now, I have some dim recollection, that gcc/clang used to replace say `std::isnan(x)` with a literal `false` even with just -O3. And that we had some very inscrutable bug reports caused by this. However, that appears not to be the case with current releases, so perhaps this has been fixed/changed?

So... I'm undecided whether the boost functions should become synonyms for the std ones, but just wanted to flag the issue.