Add the required __aeabi_* functions to libc.

The floating point functions are here rather than compiler-rt because the
libc softfloat code allows us to set the rounding mode.