Fix up error rates for cpp_bin_float: these are higher mostly down to less accurate sine and cosine than the competition :-(