scalar_to_trits does an unoptimized division by 3

[kayabaNerve]

The divisor only has two-bits, yet it performs a full 256-bit division. A tailored division should be much more efficient.