The bit pattern of a random float, favouring zero, NaN, infinities and the extremes.
float_bits :: (r: *Rng, size: s64) -> u64