math: Add inputs that yield larger errors for float type

[paul zimmermann]

       Dear all,

now that my copyright assignment is ok, you'll find on [1] a first contribution
that adds some inputs that yield larger errors for the float (binary32) type on
x86_64. These cases were generated by exhaustive search using MPFR, thus the
corresponding bounds are now definitive (on x86_64 only of course).

For those functions where no case was added, this means the previous bounds
were the largest ones.

This exhaustive search only concerns functions with one argument. It does not
cover functions like pow(x,y) or hypot(x,y) where exhaustive search would be
too expensive.

I would be grateful if someone could review that contribution.

Paul Zimmermann

[1] https://homepages.loria.fr/PZimmermann/glibc-contrib/0001-math-Add-inputs-that-yield-larger-errors-for-float-t.patch (md5sum bacf7746dd7b9cf8f8d3db12d45b932e)