cf #29
Rather than sorting the size at all we could store the string in null-padded form.
This would let us use up to the full width of the backing primative for content.
Then to determine the ncodeunits we would count backwards from the end how meany consecutive bytes were not 0x00
and subtract that from the size of the backing primative