halfagg: Expand on limit of signatures

[real-or-random]

Alternative to #16 that just improves the text a bit.

A middle way between this and #16 will be to make the limit a parameter. That is, the algorithms enforce some limit, but we leave the actual value unspecified. The more I think about this, the more I like the idea. It frees us from picking a value, but we still tell implementors how to handle the limit. The only drawback of this approach is that we'll have to pick some value in our tests in order to instantiate the algorithm. I guess, that's fine, and we can really pick whatever we want, even much lower than 2^16-1?!

What do you think?