Update floating_point.GenFloat to new Coq realization format, so as to ease tracking of the original theory.

It cannot be made part of the update-coq Makefile target due to some
peculiarities (sections and records) that are not supported.