First, have oct:float-nan-p call octi:float-nan-p so that we're
consistent with whatever is done in octi:float-nan-p.

Second, add some comment for float-nan-p for qd-reals: we're only
checking the first element for NaN.