This is the single-precision analogue of the cos function
which we have been discussing the past few weeks.

There should not be much different from the patch implementing
the double-precision version.

A namespace correction for constants in the double-precision
implementation is here. cos/cosf are together in BZ #24205, so it
should be okay that changes for both are in the same patch.

As an aside, while examining the x86_64 implementation, I found
about two-dozen files missing the FSF copyright header. They are
all the single-line files shown when you run this command:

wc sysdeps/x86_64/fpu/test-[df]*