Instead of throwing an error or zero-padding, "compress" now returns the
input bytes if there are less than or equal to "target" number of them.
I think this is logical since the goal of compress is to reduce the
complexity of the digest before making it human consumable. In this case
the complexity is already low enough to proceed.

Excess bytes are now distributed amongst the compressed bytes, instead
of being dumped into the final bit as they were before.